Censoring-Aware Models Quantify Systematic Underestimation of PFAS Occurrence in U.S. Drinking-Water Monitoring: Reproducibility compendium

Reproducibility compendium for the associated manuscript, “Censoring-Aware Models Quantify Systematic Underestimation of PFAS Occurrence in U.S. Drinking-Water Monitoring.” Version 2 retains the locked January 15, 2026 EPA UCMR 5 and SDWA inputs and the Step 1–4 production pipeline from version 1 and adds the national robustness analyses used in the revised manuscript: direct censored-likelihood benchmarking; lognormal, gamma, and Weibull marginal-family sensitivity analyses; regression-on-order-statistics and reverse-Kaplan–Meier estimability diagnostics; source-water and system-size refits; likelihood-Hessian parameter-uncertainty propagation; EPA-aligned PFOA/PFOS sampling-location benchmark-proximity analyses; prospective lower-reporting-limit scenarios; revised Figure 3; source-data tables; and SHA-256 integrity records. The standalone Step 5b–9 scripts are functionally equivalent reconstructions from preserved frozen inputs, archived outputs, retained code, and analytical documentation and were numerically verified against the archived revision results. Detected concentrations are observed, whereas concentrations below minimum reporting levels are model estimated under the stated assumptions. PFAS-CAST v1.1 is archived separately as an exploratory screening and prioritization companion and is not the national headline-estimation pipeline. Code is licensed under the MIT License; derived data products are licensed under CC BY 4.0; raw EPA and U.S. Census inputs retain their public-domain status at source.