Data audit (P1)
| file | SHA-256 | lines (wc -l) / items | notes |
|---|---|---|---|
| data/assertions.json | 2965ab3acc6f326653169b83a5345698b51fac88a556de65faf89210aea0cad4 | 6 top-level items | ids A1–A6; A1–A5 cite arxiv-2609.27732v1, A6 cites arxiv-2606.14621v1; no duplicates |
| data/README.md | 5348bd49912099735108dbbe19ae09fa72bd600cb3d12b878658e6b9806198a7 | 14 | provenance; listed extract hashes match the files |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b | 134 | matches README hash; Tables 6, 7, A19 + text |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 | 58 | matches README hash; Tables 2, 3 re-set by editor |
Units: DCR ATTs in log points of hospitalisation rate per 100,000; SEs clustered at MedStat region. RESEA: weeks, $, pp; 95% CIs and p-values printed. No date-range or tabular missing-value issues apply (text extracts). No input violates availability rules; all inputs admissible.
Anomalies (internal inconsistencies of sources; not affecting decisions unless noted):
- DCR Table A19 prints SE (0.218) for the CS-DR 5 km ATT −0.431 whereas Table 6 prints (0.210); A19 note says 640 never-treated regions while its body and Tables 6/7 say 608. A1 is decided on Table 6 (the overall-ATT table); significance at 5% holds under both SEs (p≈0.040 / 0.048, normal approx.).
- DCR abstract says fatal hospitalisations "fall by 35%" while Table 7 5 km −0.354 implies −29.8% and Sec 6.1 says "approximately 30%"; abstract says narcotics rise "by a comparable amount" while 0.390 implies +47.7% (Sec 6.1: 47%). Not used for any verdict.
- DCR text says HIV estimates are "close to zero" although 3/5 km are −0.239/−0.340 (insignificant). A2 decided on printed sign of the 1 km cell.
- DCR Figure 7 note mentions "four" sub-categories; Table 7 has five. Not used.
- RESEA text prints CIs for reductions as positive magnitudes ([1.75, 2.24]) while Table 2 prints signed [−2.24, −1.75]; equivalent. Text contains rendering defect "p ¡ 0.001" (= p < 0.001).