Data audit (P1) — 2026-10-30
| file | SHA-256 | lines (wc -l) / items | matches README hash |
|---|---|---|---|
| data/effects.json | 708c1ff0c4379c228ac6a98f577c23d0d61564f3c958fb35596266aca1f5f10b | 37 lines; 5 top-level items | n/a (not listed) |
| data/README.md | 7cf7cb385f9fb4fc0df52cfc6b0644bcbfb969a38afe2b018b864b7871187dc4 | 16 | n/a |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b | 134 | yes |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 | 58 | yes |
| data/sources/arxiv-2505.04555v3.txt | b1731a9b1c40ae91c6be2cf2e92623bb6cf86c551587936a0270356f76ae95c1 | 66 | yes |
| data/sources/arxiv-2510.03658v1.txt | f8782937eeaebd2dae0e653bbfc6aeb266c51587d09a4207bcaa3fa48a881127 | 91 | yes |
Schema: effects.json is a list of 5 objects with keys id, source, effect, reported_definition, target_definition; ids E1–E5 unique, no missing fields. Sources are plain-text extracts with provenance header; tables one row per line, " | " separators, SE on following line.
Units: DCR outcomes = log hospitalisation rate per 100,000 (ATT in log points). RESEA reemployment means in %, effects in pp. MW Table 2: elasticities unitless; '% MW changes' = 0.047 (interpreted as a proportion, see anomalies). Tossou Table 3: LPM coefficients on 0–1 probability; Mean Dep. Var. 0.261.
Anomalies (all also recorded in results.json data_issues):
- DCR abstract "fatal outcomes fall by 35%" vs Table 7 5 km -0.354 (=-29.8%); Section 6.1 says ~30%.
- DCR Table A19 SE (0.218) vs Table 6 SE (0.210) for the same -0.431; A19 note "640 regions" vs 608 in tables.
- DCR Table A19 % change last-digit mismatches for -0.362 and -0.149.
- MW Table 2 '% MW changes' 0.047 vs note "about 4%".
- Tossou % Change -0.3685 vs -0.0962/0.261=-0.3686; abstract cites Table 2 (not in extract).
- RESEA control mean is raw (not block-adjusted); abstract rounds 1.47 to 1.5 pp.
- Rendering defects (footnotes inline, 'p ¡ 0.001').
- No fatal-hospitalisation rate level (mean/baseline) printed anywhere in the DCR extract -> E4 not harmonisable.
Information-availability: no input violates the rules; only the extracts are used. No remembered values used. Exclusions: none (out/exclusions.csv).