Reproducibility package: Machine learning for municipal crop yield prediction in Chiapas, Mexico (maize, beans, coffee cherry; SIAP 2000–2023, NASA POWER/MERRA-2)

Reproducibility package for the manuscript "Machine learning para predecir rendimiento agrícola municipal en Chiapas: generalización espacial con datos climáticos y geográficos". It contains: (1) the frozen scientific package POST_TEST_V4 (evaluation code, metrics CSVs, frozen predictions, permutation and SHAP interpretability, 300-dpi figures; internal SHA-256 manifest of 52 files; package SHA-256: 6bce313ed5690d8c299715194edb4e9045822bbe8ee491540c1496afb6242bb2); (2) independent numerical replication evidence (tolerance 1e-10) of all metrics, persistence comparisons and municipality-clustered bootstrap; (3) complementary analyses (August 2026): partition descriptive table, paired bootstrap of the RMSE difference (advanced minus persistence), additional mean baselines, five-fold spatial cross-validation grouped by municipality, new figures (study-area partition map, observed-vs-predicted, methodological pipeline) and a ready-to-run script to retrieve SoilGrids soil covariates for the municipal seats. Source data are public: SIAP municipal agricultural closures (Mexico), NASA POWER/MERRA-2 daily climate, and INEGI cartography. Requirements: Python >= 3.10 with scikit-learn, xgboost, shap, pandas