An open, multi-state, depth-resolved inventory of per- and polyfluoroalkyl substances (PFAS) in soil

An open, multi-state, depth-resolved inventory of per- and polyfluoroalkyl substances (PFAS) in soil, harvested from public U.S. agency site-investigation reports with a vision–language-model (VLM) extraction pipeline. Companion dataset to a manuscript submitted to Environmental Science & Technology and to the SWATGenX national groundwater-lithology inventory (Zenodo 10.5281/zenodo.20973076). Contents: 34,092 PFAS soil measurement records (33,866 with a canonical analyte) resolving to 7,788 distinct soil samples across 6,342 borings, drawn from 144 public report sources spanning 41 U.S. states and 73 distinct PFAS. 22,718 records carry an explicit, sanity-passed depth interval. Released as CSVs plus a single SQLite file, the full extraction/QA pipeline code, a README, a data dictionary, and a per-report provenance registry linking every record to its original public document. Soil-only (no groundwater, surface water, or private drinking-water-well data). Ten QA gates and an independent accuracy audit (92.7% value-presence) are documented. Units: ng/g dry weight. License: data CC-BY-4.0; code MIT.