# Data audit — F4-conf-03

| file | SHA-256 | lines (wc -l) / items | matches README/bundle hash |
|---|---|---|---|
| data/effects.json | 708c1ff0c4379c228ac6a98f577c23d0d61564f3c958fb35596266aca1f5f10b | 5 top-level items (list) | yes (bundle.json) |
| data/README.md | 7cf7cb385f9fb4fc0df52cfc6b0644bcbfb969a38afe2b018b864b7871187dc4 | 16 lines | yes (bundle.json) |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b | 134 lines | yes |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 | 58 lines | yes |
| data/sources/arxiv-2505.04555v3.txt | b1731a9b1c40ae91c6be2cf2e92623bb6cf86c551587936a0270356f76ae95c1 | 66 lines | yes |
| data/sources/arxiv-2510.03658v1.txt | f8782937eeaebd2dae0e653bbfc6aeb266c51587d09a4207bcaa3fa48a881127 | 91 lines | yes |

Schema: effects.json is a list of 5 objects {id, source, effect, reported_definition, target_definition}; ids E1–E5 unique, no missing fields, no duplicates. Sources are plain-text extracts with tables as ` | `-separated rows, SE rows following coefficient rows.

Units: DCR (2609.27732) ATTs in log points of rate per 100,000; RESEA (2606.14621) means in %, effects in pp; MW (2505.04555) elasticities (dimensionless), "% MW changes" printed as 0.047 (proportion); Tossou (2510.03658) LPM coefficients on 0–1 scale, Mean Dep. Var. 0.261.

## Anomalies found (also in results.json `data_issues`)
1. DCR abstract says fatal hospitalisations "fall by 35%", but Table 7 (5 km) prints -0.354 → 100(e^-0.354 −1) = −29.8%, and Sec. 6.1 says "approximately 30%". Table value used.
2. DCR Table A19 CS-DR SE (0.218) differs from Table 6 SE (0.210) for the same −0.431 estimate; A19 note says 640 never-treated regions vs 608 in the table rows. Point estimates unaffected.
3. DCR Table A19 printed % change for SDiD (−30.3%) and Stacked DiD (−13.9%) differ by ~0.1 from 100(e^b−1) (−30.37%, −13.84%). Not used.
4. MW Table 2 prints "% MW changes | 0.047" while its note says "about 4% minimum wage increase". Task instructs use of the table value; 0.047 interpreted as a proportion (4.7%).
5. Tossou Table 3 col (4) "% Change −0.3685" vs −0.0962/0.261 = −0.36858 (rounding of the printed mean). Text cites Table 2 (−0.0855/0.455), not in extract. Footnote markers render as "^9^9 9" (rendering defect, ignored).
6. RESEA control mean 75.0% is the unadjusted aggregate control-group mean; the ATE is block-(week-)adjusted. Used as the task-specified denominator.
7. E4 needs the level of the fatal drug-related hospitalisation rate (per 100,000) for treated regions; no such level is printed in the extract.

No input violates the information-availability rules (all are the designated extracts). No rows excluded; all 5 items analysed.