Data and code: Vadose-zone control of PFOS delivery to groundwater: explicit unsaturated-zone transport with solid-phase and air–water interfacial sorption

PFAS vadose end-members: data, code and model runs (version 3) This version accompanies the revised manuscript "Vadose-zone control of PFOS delivery to groundwater: explicit unsaturated-zone transport with solid-phase and air–water interfacial sorption" (CONHYD-D-26-01095, revision). Versions 1 and 2 remain available under the same concept DOI. CHANGES.md lists what changed since version 2. Contents pfas-vadose-endmembers-v3.zip: site data and processing code. data/wurtsmith_gw_pfas_georef.csv: the Wurtsmith groundwater PFAS well table, updated since submission. code/parse_wurtsmith_gw.py: the parser that writes it, with its check code/test_table5_known_cells.py. The Gabreski, surface-water and depth-to-water data and the builders carried from version 1. q616-patched-engine-package.tar.gz: the patched MODFLOW 6 engine as a patch series against upstream MODFLOW 6 073acdc7 (rebuild instructions in its README.md); per-run provenance (runs/<name>/SOURCE.txt, SHA256SUMS); and the acceptance record. q624-patched-engine-data.tar.gz: the input decks and result files of the patched-engine runs first reported with the revision. run__<run>__results.tar.gz: per model run the revised manuscript reports or cites, its results archive (result files, solver listings, and the input deck where the runner recorded one). A run with no results archive (a copied baseline, or an attempt kept for its listings) is run__<run>.tar.gz. The package README (q616-deposit-v2/README.md inside q616-patched-engine-package.tar.gz) maps each run to its engine version and its role. The input decks of the three Stegman closure-sensitivity runs are run__stegman_closure_{A,A2,B}__input_decks.tar.gz. A2 re-ran A's two R-A arms on the same decks, with a longer time limit for one of them. The matched pair at Wurtsmith and its sensitivity runs, each with its expected result stated before it ran: run__m1_postclosure_2026-09-30__results.tar.gz, run__m1r_robustness_am1_2026-09-30__results.tar.gz and run__m1r_robustness_am2_2026-09-30__results.tar.gz. Every run's input deck is inside its results archive. run-records.tar.gz holds each run's status and run record (instance type, billed time). Engine binaries are not included; they rebuild from the patch series. Stegman coupled-run package, carried from version 2 (MODFLOW 6.7.0, as submitted); stegman_scripts.zip with its paths rewritten (below). The revised manuscript does not report these runs' numbers; it reports the converged pair in run__stegman_sorbing_pair__results.tar.gz. The package: stegman_decks.zip, stegman_listings.zip, stegman_scripts.zip; stegman_vadose_gwt.ucn and stegman_contrast_gwt_partial.ucn; p4_wells.csv, the 846-well table behind the submitted version's Stegman well statistics (well records from Michigan EGLE's public site-monitoring inventories). The revised manuscript withdraws those statistics: no converged run reaches 2018–2021. revision-analysis-2026-09-30.tar.gz: the plans (each stating its expected results before its run), builders, scorers, smoke tests and result records of those three runs; the record of every derived number the manuscript quotes that a run's own scorer does not print (revision/derived_numbers_r2.py and its output); the instruments behind the well-to-grid counts (revision/rerun-controls/); and the script behind every table and figure of the manuscript and its supplement: the screen (screen_first_arrival.py, with the priors it imports), the transit table, the study-area map, the matched-pair, Gabreski, Stegman Creek and conceptual figures, the plume maps and sections (make_gis_figures.py, make_3d_plume.py, with the Stegman section's drawing record) and the supplement's tables. REPRODUCIBILITY_CHECKLIST.md: for each result, the file in this version, the step, and the check its output must pass (written for this version). SHA256SUMS: a checksum for every other file in this version, listed by file name. With all files downloaded into one folder, sha256sum -c SHA256SUMS checks them. Engine versions Two engine versions are used, and no printed quantity pairs results from different versions: - Patched build 6.9.0.dev0+51d96cadd: the Wurtsmith and Gabreski results, the joint source sweeps, the Stegman paired experiment (the converged fixed-step pair and its sorbing arm, in run__stegman_sorbing_pair__results.tar.gz), the sorption sensitivity runs, and the matched pair at Wurtsmith with its time-step, advection-scheme and interfacial-term runs. - Released MODFLOW 6.7.0: the submitted version's Stegman coupled-run package, the source-footprint sweep, and the run behind the submitted well classification. The revision does not report the Stegman package's numbers (its twin did not converge in its 25th stress period) or the classification (no converged run of that configuration). Both are kept as the record of what was submitted. What this version does not settle Where the manuscript reports an item, its section is named; the others are records of this version only. - Two engine versions. The revision's Wurtsmith, Gabreski, joint-source and Stegman sorbing-column results are the patched build 6.9.0.dev0+51d96cadd; the submitted version's results are released MODFLOW 6.7.0 (Engine versions, above). The two are different releases, so a difference between a revised and a submitted value is not by itself an effect of the patch: each reported sorption effect is measured against the same source tree with sorption disabled (manuscript, Section 2.3). - A registered sign test did not hold as written. We registered, before the runs, that all three layers of the Wurtsmith plume-core column would fall under sorption. In 2017 the shallow layer rises under solid-phase sorption alone, from 7.9 to 11.0 ng/L, before falling when the air-water interfacial term is added. The rewritten manuscript reports the deep layer's decomposition only. - A registered interval missed at one site. The ratio of arrival times with and without the air-water interfacial term was registered to fall in 1.55–1.65 at Wurtsmith and 1.59–1.62 at Gabreski. It is 1.61 at Wurtsmith, inside its interval, and 1.65 at Gabreski, outside its interval by 0.03; the interval was not re-derived after the read (manuscript, Section 2.3). - Two predictions of the matched pair did not hold as written. The deep-layer contrast was predicted to hold at the peak, and it holds only at one date (2017), because the explicit run's deep layer is still rising in 2024. The advection-scheme prediction did not hold: TVD in place of upstream weighting moves the deep 2017 value by 11%, a sensitivity the manuscript reports with its size (manuscript, Section 3.1). - One arm is built and run but not read. decks/v3_wurtsmith/defective_deck and results/v3_wurtsmith/defective_deck in q624-patched-engine-data.tar.gz have no registered expectation: the arm exists because the builder's arm table is the same for both sites. It is deposited for completeness and deliberately not read, so no number from it is quoted. Paths Paths that named another project's folder are shown as <research-root>: in stegman_scripts.zip and in the package's build script. Nothing else in those files changed. Our build scripts and records keep the absolute paths of our servers, where the work ran. The code a reader runs, code/ in pfas-vadose-endmembers-v3.zip, takes its inputs from one folder, VADOSE_ROOT; that archive's README lists the inputs it does not include. License: CC-BY-4.0, as versions 1 and 2.