# Data audit (P1)
| file | SHA-256 | lines (wc -l) | items | matches README hash |
|---|---|---|---|---|
| data/README.md | c6d375725356176fa7600ab663d9306c5f37c1097b619792bcbed166ccc792b2 | 17 | — | n/a |
| data/reference_list.json | 7ab6cbf2606f90f3174215214feddd73533d29cda4befdc77ffdb595193ab5da | — | 5 top-level items (R1–R5), keys id/entry/cited_for/source_file, no missing values, no duplicate ids | n/a |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d2f78f2af2ca2cf69b59a61fcd7ac5e10ea29e8ddcd67917bf07d6a8b | 134 | — | yes |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0a144d9d5a3916dc641f15b6029f6c68b5e574ad9074e7df1aeb0a238 | 58 | — | yes |
| data/sources/arxiv-2603.03260v2.txt | 3cb9a4f3574e7b2272fe19cde165a745543883f86beb9342f6fc0ab744c31551 | 107 | — | yes |
| data/sources/arxiv-2505.04555v3.txt | b1731a9b1c40ae91c6be2cf2e92623bb6cf86c551587936a0270356f76ae95c1 | 66 | — | yes |
| data/sources/arxiv-2510.03658v1.txt | f8782937eeaebd2dae0e653bbfc6aeb266c51587d09a4207bcaa3fa48a881127 | 91 | — | yes |

Every reference's source_file exists. Each extract has a "Title:" and "Bibliographic record:" header line and a body "# " title line; header title == body title in all five. Units: regression coefficients in log points (2609, 2603), dollars/pp/weeks (2606), elasticities (2505), probabilities (2510).

## Anomalies found (recorded as data_issues in results.json)
1. 2609.27732 (Armendariz): CS-DR 5 km ATT −0.431 has SE (0.210) in Table 6 but (0.218) in Table A19 (same estimator/sample, 17,250 obs). Table A19 note says 640 never-treated regions vs 608 in its own body and in Tables 6/7. Abstract says fatal hospitalisations fall by 35% and narcotics rise "by a comparable amount", whereas Table 7 (−0.354, +0.390 log points) and Sec. 6.1 give ≈30% and 47%. None of these touch the elements cited in R1 (both SEs imply p<0.05: t=2.05 and 1.98, both starred **).
2. 2505.04555 (Kanayama et al.): Table 2 own-wage elasticity −0.732 is printed with SE (0.091), identical to the SE of "Affected employment"; a delta-method approximation from the printed components (ignoring covariance) gives ≈0.42. Possible transcription error in the source; R4 reproduces the printed value, so it is "stated in the extract".
3. 2603.03260 (Liu & Zhao): Table 3 header row garbled in rendering; extract editor supplies column order. Used as given.
4. 2510.03658 (Tossou): footnote markers rendered inline ("percent^9^9 9"); footnote 11 percentages detached from the text. Not relevant to R5's cited cells.
5. 2606.14621 (Li et al.): Tables 2–3 were re-set by the extract editor (declared in header); "p ¡ 0.001" rendering defect in text. Values used as printed.

## Availability
No input violates the information rules; only extracts are used. Headers give submission dates for the year check.