# Data audit — F4-conf-03

All SHA-256 values computed with `shasum -a 256` match both data/README.md and bundle.json.

| file | SHA-256 | lines (wc -l) / items | content |
|---|---|---|---|
| data/effects.json | 708c1ff0c4379c228ac6a98f577c23d0d61564f3c958fb35596266aca1f5f10b | 5 top-level items (list) | fields id, source, effect, reported_definition, target_definition; ids E1..E5 unique, no missing fields |
| data/README.md | 7cf7cb385f9fb4fc0df52cfc6b0644bcbfb969a38afe2b018b864b7871187dc4 | 16 lines | provenance table |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b | 134 lines | DCR paper: abstract, Tables 6, 7, Sec 6.1, Table A19 |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 | 58 lines | RESEA: abstract, Table 2, Table 3 (re-set by extract editor) |
| data/sources/arxiv-2505.04555v3.txt | b1731a9b1c40ae91c6be2cf2e92623bb6cf86c551587936a0270356f76ae95c1 | 66 lines | MW Japan: abstract, Appendix A.1 Table 2 |
| data/sources/arxiv-2510.03658v1.txt | f8782937eeaebd2dae0e653bbfc6aeb266c51587d09a4207bcaa3fa48a881127 | 91 lines | Benin C-section: abstract, Sec 4.1, Table 3 |

Units: DCR outcomes log rate per 100,000 (ATT log points); RESEA reemployment % (means) / pp (effects); MW elasticity unitless, "% MW changes" printed as a proportion (0.047); Benin LPM coefficient on 0-1 scale.

No duplicates or missing values in effects.json. No tabular row-level data, so no rows are excluded (see out/exclusions.csv).

## Anomalies found (also in results.json `data_issues`)
1. DCR abstract: fatal hospitalisations "fall by 35%"; Table 7 5 km fatal ATT -0.354 implies 100(exp(-0.354)-1) = -29.8%, matching Sec 6.1's "approximately 30%". Abstract inconsistent.
2. DCR: Table A19 CS-DR SE (0.218) differs from Table 6 5 km SE (0.210) for the same -0.431 estimate; A19 note says 640 never-treated regions vs 608 in tables.
3. DCR Table A19 "% change" rows: -0.362 -> -30.37% (printed -30.3%), -0.149 -> -13.84% (printed -13.9%): small rounding inconsistencies (may reflect unrounded coefficients).
4. DCR: no baseline / counterfactual level of fatal hospitalisation rate printed anywhere in extract -> E4 not harmonisable.
5. MW Table 2 note says "about 4% minimum wage increase" while the table prints "% MW changes 0.047" (4.7%). Target definition says use the size printed in the table -> 0.047 used.
6. Benin text cites "table 2" (-0.0855, mean 0.455; Table 2 not in extract) and also misinterprets the relative change ("for over 100 newborns ... 18 are saved"); abstract calls the relative change a "percentage change". Table 3 col (4): -0.0962/0.261 = -0.3686 vs printed -0.3685 (rounding, plausibly unrounded coefficient). Footnote markers render as "^9^9 9" (rendering defect).
7. RESEA: control mean 75.0% is a raw (unadjusted) group mean whereas ATE adjusts for randomisation week; abstract rounds ATE to 1.5 pp; "p ¡ 0.001" rendering defect.

## Information-availability
No input violates the rules. All quantities used are printed in the cited extract. The only unavailable quantity (baseline fatal rate for E4) is not substituted.