Data audit
| file | SHA-256 | lines (wc -l) | matches README/bundle |
|---|---|---|---|
| data/README.md | c6d375725356176fa7600ab663d9306c5f37c1097b619792bcbed166ccc792b2 | 17 | yes (bundle.json) |
| data/reference_list.json | 7ab6cbf2606f90f3174215214feddd73533d29cda4befdc77ffdb595193ab5da | 5 top-level items | yes (bundle.json) |
| data/sources/arxiv-2609.27732v1.txt | ee2d444d…a8b | 134 | yes |
| data/sources/arxiv-2606.14621v1.txt | 27c463c0…238 | 58 | yes |
| data/sources/arxiv-2603.03260v2.txt | 3cb9a4f3…55 | 107 | yes |
| data/sources/arxiv-2505.04555v3.txt | b1731a9b…5cb | 66 | yes |
| data/sources/arxiv-2510.03658v1.txt | f8782937…127 | 91 | yes |
Schema: reference_list.json = list of 5 objects {id, entry, cited_for, source_file}; ids R1–R5 unique, no missing fields, every source_file exists. Each extract has a provenance header (Title, Authors, Bibliographic record with submission dates) then the body.
Dates: R1 source v1 23 Sep 2026; R2 v1 12 Jun 2026; R3 v1 3 Mar 2026, v2 2 Apr 2026; R4 v1 7 May 2025, v3 16 Mar 2026; R5 v1 4 Oct 2025.
Anomalies found (internal to sources)
- arxiv-2609.27732v1: Table 6 5 km ATT SE (0.210) vs Table A19 CS-DR column SE (0.218) for the same specification; Table 6 says 608 never-treated control regions, A19 note says 640. Abstract says fatal outcomes fall by 35% whereas Section 6.1 says ~30% (Table 7: -0.354 log points).
- arxiv-2505.04555v3: SE of own-wage elasticity (0.091) identical to SE of "Affected employment"; table note says "about 4%" MW increase while "% MW changes" row is 0.047.
- arxiv-2510.03658v1: Table 3 "% Change" col (4) is -0.3685 while -0.0962/0.261 = -0.3686 (rounding); abstract's headline (0.0855, 18.79%, mean 0.455) refers to a different outcome/table (Table 2, omitted).
- arxiv-2603.03260v2: Table 3 header row garbled (editor note supplies column names); used as documented.
- arxiv-2606.14621v1: Tables re-set by extract editor (documented); "p ¡ 0.001" rendering glyph in Section 3 text.
Information-availability
No input violates the rules; only the extracts are used. No exclusions are applied (5 references in, 5 analysed).