# Data audit (P1)
| file | SHA-256 | lines (wc -l) / items | notes |
|---|---|---|---|
| data/assertions.json | 2965ab3acc6f326653169b83a5345698b51fac88a556de65faf89210aea0cad4 | 32 lines; 6 top-level items | ids A1–A6, unique; A1–A5 cite arxiv-2609.27732v1, A6 cites arxiv-2606.14621v1; no missing fields |
| data/README.md | 5348bd49912099735108dbbe19ae09fa72bd600cb3d12b878658e6b9806198a7 | 14 | provenance; lists extract SHA-256s |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b (matches README) | 134 | DCR paper: abstract, Sec. 6, Tables 6, 7, A19. Units: log points of hospitalisation rate per 100,000; SEs clustered by MedStat region |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 (matches README) | 58 | RESEA RCT: Table 2, Sec. 3 text, 3.1, Table 3. Units: $, pp, weeks |

Schema: tables are one row per line, ` | ` separated; SE rows follow coefficient rows (DCR). No duplicates among assertions.

Anomalies (also in results.json `data_issues`):
- DCR: CS-DR 5 km ATT SE 0.210 (Table 6) vs 0.218 (Table A19).
- DCR: A19 note says 640 control regions vs 608 elsewhere.
- DCR: narrative % (fatal 35% in abstract vs ~30% in 6.1; HIV "close to zero" vs -0.340 at 5 km).
- RESEA: text gives CI as positive magnitudes; "p ¡ 0.001" rendering defect.
- Both extracts are partial; RESEA tables re-set by editor.

Information availability: no input violates the rules; only the two extracts are used as evidence. assertions.json is the object of evaluation, not evidence.
No rows excluded (see out/exclusions.csv).