Data audit (P1)
All SHA-256 values below were computed with shasum -a 256 and match both data/README.md and bundle.json.
| file | SHA-256 | lines (wc -l) / items | content |
|---|---|---|---|
| data/README.md | c6d37572…792b2 | 17 lines | provenance table |
| data/reference_list.json | 7ab6cbf2…5da | 5 top-level items (id, entry, cited_for, source_file) | references R1–R5 |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d…8a8b | 134 | Armendariz; v1 23 Sep 2026 |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0…b0a238 | 58 | Li et al.; v1 12 Jun 2026 |
| data/sources/arxiv-2603.03260v2.txt | 3cb9a4f3…51589 | 107 | Liu & Zhao; v1 3 Mar 2026, v2 2 Apr 2026 |
| data/sources/arxiv-2505.04555v3.txt | b1731a9b…95c1 | 66 | Kanayama et al.; v1 7 May 2025, v2 27 May 2025, v3 16 Mar 2026 |
| data/sources/arxiv-2510.03658v1.txt | f8782937…1127 | 91 | Tossou; v1 4 Oct 2025 |
No missing files, no duplicate reference ids, every source_file exists. Units: coefficients in the units stated in each extract (log points, $, pp, weeks, probabilities).
Anomalies found (source-internal; none changes the inputs of the checks)
- arxiv-2609.27732v1: Table 6 5 km SE (0.210) vs Table A19 CS-DR SE (0.218) for the same ATT −0.431; Table A19 note says 640 never-treated regions, tables say 608; abstract says fatal outcomes fall 35% and narcotics rise "by a comparable amount", Section 6.1 says ~30% and 47% (exp(−0.354)−1 ≈ −29.8%, exp(0.390)−1 ≈ +47.7%).
- arxiv-2606.14621v1: "p ¡ 0.001" rendering defect; Tables 2–3 re-set by extract editor (declared).
- arxiv-2603.03260v2: Table 3 header garbled in HTML; editor note gives column order (used).
- arxiv-2505.04555v3: printed SE of own-wage elasticity (0.091) equals the SE of 'Affected employment'; note says "about 4%" MW increase vs '% MW changes' 0.047. Not decisive for the citation check.
- arxiv-2510.03658v1: abstract headline (0.0855, 18.79%, mean 0.455; Table 2, omitted) differs from Table 3 (0.0962, 36.85%, mean 0.261) — different outcome definitions; footnote 11 garbled (three percentages for five wealth categories); footnote markers rendered as "^9^9 9".
Information-availability
No input violates the rules: all evidence is the extracts themselves. HTML rendering dates are explicitly excluded by the extract headers; only the "Bibliographic record" line is used for years.
Exclusions
None. All 5 references analysed (see out/exclusions.csv).