# Data audit (P1)

Schema: plain-text extracts; tables one row per line, " | " separators; DCR regression tables have SE line (parentheses) after each coefficient row. RESEA Tables 2–3 re-set one row per outcome with "estimate; p=...; 95% CI: [lo, hi]" cells. Units: DCR ATTs in log points (outcome log hospitalisation rate per 100,000); Observations = region-years. RESEA wages annualised USD, reemployment % / pp, weeks on UI in weeks. Date ranges: DCR openings 1998–2022; RESEA RCT Feb 2022 – Sep 2023. Missing / duplicates: no missing cells in requested rows. Duplicates: RESEA Table 3 col 1 duplicates Table 2 ATE column (identical, checked). DCR Table A19 CS-DR column duplicates Table 6 5 km ATT but with a different SE (0.218 vs 0.210) — anomaly. Anomalies: (1) SE 0.210 (Table 6) vs 0.218 (Table A19) for same estimate; (2) 608 vs 640 control regions (A19 note); (3) abstract "fatal fall by 35%" vs Sec 6.1 "~30% at 5 km"; text says HIV "close to zero" though 5 km point is -0.340; (4) RESEA text gives weeks-on-UI CI as unsigned magnitudes; (5) "p ¡ 0.001" rendering defect. All recorded in results.json data_issues. Information-availability: no input violates the rules; only the two extracts are used; README used for provenance/hash verification only. No exclusions applied (this is an extraction task; no rows filtered).