Data audit — F4-conf-01
| file | SHA-256 | lines (wc -l) | matches README/bundle hash |
|---|---|---|---|
| data/README.md | 16ebdff389eeaee9508fe81c4edc43690f31f3463c469680c814134086ef9929 | 10 | yes (bundle.json) |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b | 134 | yes |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 | 58 | yes |
Schema: plain-text extracts; tables one row per line, cells separated by " | "; in DCR tables the line after a coefficient row holds SEs in parentheses. RESEA Tables 2–3 re-set one row per outcome; each ATE cell = "estimate; p=..; 95% CI: [lo, hi]". Date range: N/A (published tables; extracts retrieved 2026-10-30T03:23:18Z). Units: DCR ATTs in log points (outcome = log hospitalisation rate per 100,000); observations = region-years. RESEA wages USD/yr (printed with $ and thousands separators), reemployment pp, weeks on UI in weeks. Missing values: none in the requested cells. No exact p-values are printed in DCR Tables 6/7 (stars only). No SE is printed for any RESEA estimate (CIs and p-values only). Duplicates: Table 6 5 km ATT duplicated in Table A19 (CS-DR column) with a DIFFERENT SE (0.210 vs 0.218); RESEA Table 2 ATE column duplicated in Table 3 "Difference in means" (identical values).
Anomalies / internal inconsistencies
- DCR: SE of the 5 km CS-DR ATT is (0.210) in Table 6 but (0.218) in Table A19 for the same estimate (-0.431**, 17,250 obs). Handling: key names Table 6 → 0.210 used; flagged in data_issues.
- DCR: control regions stated as 608 (Tables 6, 7, Fig. 6 note) but "640 regions" in the Table A19 note. Observation counts equal 25 × (treated + 640): 16,175 = 25×647, 16,875 = 25×675, 17,250 = 25×690; with 608 controls they are not integer multiples of the region count. Handling: Observations reported as printed (17250); flagged (checked numerically in analysis.py).
- DCR narrative: abstract says fatal hospitalisations "fall by 35%", Section 6.1 says "approximately 30% at 5 km" (exp(-0.354)-1 ≈ -29.8%); text says HIV estimates "close to zero" although the 5 km ATT is -0.340 (not starred). Narrative only; no extracted value affected.
- RESEA narrative gives the weeks-on-UI CI as magnitudes "[1.75 weeks, 2.24 weeks]" vs signed [-2.24, -1.75] in Table 2. Handling: Table 2 signed values used.
- RESEA rendering defect: "p ¡ 0.001" (line 39) for "p < 0.001". No extracted value affected.
- RESEA terminology: Table 2 column "Average treatment effect estimate"; Table 3 caption calls the same numbers ITT estimates. Noted only.
Information-availability
Only the two extracts (and README for provenance) are used. No input violates the rules. Table A19 and Table 3 are used only for cross-checks, never as sources for keys.