# Data audit (P1)
All SHA-256 values recomputed with `shasum -a 256` and match data/README.md and bundle.json.

| file | sha256 | wc -l | content |
|---|---|---|---|
| data/README.md | c6d37572…c792b2 | 17 | provenance table |
| data/reference_list.json | 7ab6cbf2…5da | 32 (5 top-level items) | R1–R5 {id, entry, cited_for, source_file} |
| sources/arxiv-2609.27732v1.txt | ee2d444d…8b | 134 | Armendariz; v1 23 Sep 2026 only; abstract, Sec 6, Tables 6,7, A19 |
| sources/arxiv-2606.14621v1.txt | 27c463c0…238 | 58 | Li et al.; v1 12 Jun 2026 only; abstract, Sec 3, Tables 2,3 (re-set) |
| sources/arxiv-2603.03260v2.txt | 3cb9a4f3…551 | 107 | Liu & Zhao; v1 3 Mar 2026, v2 2 Apr 2026; abstract, Sec 3, Tables 2,3, Sec 5 |
| sources/arxiv-2505.04555v3.txt | b1731a9b…95c1 | 66 | Kanayama et al.; v1 7 May 2025, v3 16 Mar 2026; abstract, App A.1 Table 2 |
| sources/arxiv-2510.03658v1.txt | f8782937…127 | 91 | Tossou; v1 4 Oct 2025 only; abstract, Sec 4.1, Table 3 |

No missing files, no duplicate reference ids, all 5 source_file paths exist. No availability-rule violations in inputs (all evidence lies in the extracts; README duplicates header info).

## Anomalies found inside sources (recorded as data_issues, none decisive for a finding check)
1. Armendariz: Table 6 5 km ATT SE (0.210) vs Table A19 CS-DR SE (0.218) for the same −0.431 estimate; Table 6 says 608 control regions vs A19 note "640 regions". Abstract says fatal outcomes fall 35%, Sec 6.1 says ~30%; exp(−0.354)−1 = −29.8%.
2. Kanayama: own-wage elasticity SE (0.091) identical to the affected-employment SE (0.091); point estimate −0.732 = −0.268/0.366 checks out, SE not reproducible from extract. Note says "about 4%" MW increase vs "% MW changes 0.047".
3. Liu & Zhao: Table 3 header garbled (editor reconstructed column order); τ=3 coefficient not reported in text.
4. Tossou: abstract headline (0.0855, 18.79%, refers to Table 2 not in extract, mean 0.455) differs from Table 3 (binary outcome, 0.0962, 36.85%, mean 0.261) — different outcome definitions, not a contradiction per se. 0.0962/0.261 = 0.36858 (printed −0.3685).
5. Li et al.: Tables 2–3 re-set by the extract editor (values stated verbatim).