Data audit (P1)
| file | SHA-256 | lines (wc -l) | matches README/bundle hash |
|---|---|---|---|
| data/README.md | 16ebdff389eeaee9508fe81c4edc43690f31f3463c469680c814134086ef9929 | 10 | matches bundle.json |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b | 134 | yes |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 | 58 | yes |
Schema: plain text; tables one row per line with " | " separators. DCR extract: Table 6 (cols 1/3/5 km; rows ATT, SE line, % change, Treated regions, Control regions, Observations), Table 7 (rows Mental/Behavioral, Hepatitis, HIV, Narcotics, Drug-related (fatal), each followed by SE line; Treated/Control regions), Table A19 (5 km, 4 estimators). RESEA extract: Table 2 (6 outcomes x Treatment|Control|T-C difference|ATE, each effect cell "estimate; p=..; 95% CI: [lo, hi]"), Table 3 (3 outcomes x 3 estimators). Units: log points (DCR), USD/yr, pp, weeks (RESEA). No missing cells among requested values; no duplicate rows.
Anomalies / internal inconsistencies
- DCR Table 6 5 km ATT SE printed (0.210) while Table A19 CS-DR column (same estimator, same 5 km buffer, same ATT -0.431**, same 17,250 obs) prints (0.218). Handling: use Table 6 as the key requires; record as data issue.
- DCR control-region count: Table 6/7 and Fig. 6 note say 608 never-treated regions; Table A19 note says 640. Observations in Table 6 equal (640 + treated) x 25 for all three columns (16,175 = 647x25; 16,875 = 675x25; 17,250 = 690x25), not (608 + treated) x 25. Handling: report Observations as printed; flag. (25 = years 1998-2022 is my inference, not stated in the extract.)
- DCR prose: abstract says fatal outcomes "fall by 35%" whereas Table 7 5 km fatal ATT -0.354 corresponds to ~-30% (Section 6.1 says ~30%); prose says HIV estimates "close to zero" while 5 km HIV ATT is -0.340 (insignificant). Handling: tables used; noted.
- RESEA prose gives the weeks-on-UI CI as magnitudes "[1.75 weeks, 2.24 weeks]" while Table 2 prints [-2.24, -1.75]. Handling: Table 2 used, sign as printed.
- RESEA rendering defect: "p ¡ 0.001" in Section 3 text (should be p < 0.001). Not used for any key.
- RESEA Tables 2-3 re-set by extract editor (stated in header); values declared verbatim. Table 3 col 1 duplicates Table 2 ATE column.
Information-availability
No input violates the rules; only the two extracts used. README used only for provenance. No exclusions/filters applied (exclusion ledger has a single no-op row).