1# F4-conf-04 — Verification of a five-item reference list against source extracts23## Question4For each of five references (R1–R5) in a draft manuscript, I checked against the provided arXiv extract whether (i) the title matches the document, (ii) the year matches the first-submission year or the year of the reproduced version, and (iii) every part of the attributed finding ("cited_for") is stated in, or follows arithmetically from, the extract. A reference counts as unverifiable if it fails any of the three checks.56## Data and exclusions7- `data/reference_list.json` (5 items) and five extracts under `data/sources/` (66–134 lines each); `data/README.md` (17 lines) was read for provenance.8- The only evidence used was the extracts (provenance header plus body). Nothing was excluded. I did not use any external or remembered bibliographic information.910## Method11`out/analysis.py` does the following:121. **Title check:** parses the title and year from each entry. It then compares the title with both the header `Title:` line and the body `# ` title line, after normalising case, punctuation, diacritics and "&"/"and".132. **Year check:** extracts every submission year from the "Bibliographic record" line. The year passes if the entry year equals the first or the last of those years.143. **Finding check:** locates the decisive table rows and text sentences and prints them. Each element of `cited_for` is compared with the printed cell, using exact strings with the same decimals. Where an element can be derived, the arithmetic is recomputed: exp(−0.431)−1 = −35.0%; −0.268/0.366 = −0.732; −0.0962/0.261 = −0.3686, printed as −0.3685. The finding checks are applied by me and coded explicitly in the script, which asserts the exact cell values.1516## Results1718| id | title | year | finding | locator |19|---|---|---|---|---|20| R1 Armendariz | ✓ | ✓ (2026) | ✓ | arxiv-2609.27732v1, Table 6, column 5 km (ATT −0.431**, SE (0.210), −35.0%); abstract |21| R2 Li et al. | ✓ | ✓ (2026) | ✓ | arxiv-2606.14621v1, Table 2, ATT column (Wage $1,153 [166, 2,141]; Reemployment 1.47 pp; Weeks on UI −1.99); abstract (N = 23,549, preregistered RCT) |22| R3 Liu & Zhao | ✗ | ✗ (entry 2024; record 2026) | ✗ | arxiv-2603.03260v2, Sec. 3.2 stacked coefficients (0.005 at τ=0 … 0.037/0.034 at τ=4/5); Table 3 row Clean-onset threshold-2 |23| R4 Kanayama et al. | ✓ | ✓ (2025 = v1 year) | ✓ | arxiv-2505.04555v3, App. A.1, Table 2 (−0.388 (0.132); −0.732 (0.091)); abstract (Japan's spot labor market) |24| R5 Tossou | ✓ | ✓ (2025) | ✓ | arxiv-2510.03658v1, Table 3, column (4) (−0.0962 (0.0129), % change −0.3685); Sec. 4.1 says column (4) is the preferred specification |2526**Unverifiable references: ["R3"]. n_verified = 4.**2728R3 fails all three checks:29- **Title:** the entry's title ("Restaurants, Cafés and Neighbourhood Change: Commercial Entry and House Prices in Greater London") is a different title from the document's ("Does Entry of Food-and-Drink Establishments Raise Local House Prices? Event-Study Evidence from London").30- **Year:** 2024 matches neither submission date (3 Mar 2026 and 2 Apr 2026).31- **Finding:** the attributed finding contradicts the source. The source reports a coefficient of about 0.5% in the event year that rises gradually to 3.4–3.7% by years four and five, and states explicitly that "prices do not jump immediately". It does not report a 3.4–3.7% jump in the entry year followed by no further appreciation.3233## Uncertainty34No sampling or estimation is involved; the checks are deterministic string and arithmetic comparisons. The remaining uncertainty is about judgement and source quality:35- **R1:** the source is internally inconsistent. The same 5 km estimate, ATT −0.431**, has SE 0.210 in Table 6 but 0.218 in Table A19. I counted the cited SE as stated because it matches the main results table exactly. The 5% significance claim holds under either SE (|t| ≈ 2.05 or 1.98). A stricter reader might regard the uncertainty statement as unsettled.36- **R4:** the printed SE of the own-wage elasticity, (0.091), is identical to the SE of "Affected employment". That is implausible for a ratio estimate whose denominator has SE 0.171, and suggests a possible duplication in the source. The cited value nevertheless equals the printed value.37- **R5:** the "36.85%" in the citation equals the printed −0.3685. Recomputing it from the rounded cells gives −0.3686, a difference attributable only to rounding.3839## Limitations and open issues40- The extracts are partial, so material outside them (other tables, other versions) could not be consulted, by design.41- Other source-level problems were detected and logged in `data_issues`; none of them changes a verdict:42 - R1: the Table A19 note says 640 control regions, while the table rows say 608. The abstract says fatal hospitalisations fall by 35%, but Section 6.1 says about 30%; Table 7 gives exp(−0.354)−1 = −29.8%.43 - R2: the text contains the rendering defect "p ¡ 0.001".44 - R3: the Table 3 header is garbled; the column names come from the extract editor's note.45 - R5: the abstract headline (0.0855, 18.79%) refers to a Table 2 that is omitted from the extract.46- Author formatting was not checked, per the task rules.47