F4-conf-02 — Are the six assertions supported by their cited sources?
Question
For each of six assertions in a draft review (A1–A5 citing Armendariz 2026, arXiv:2609.27732v1; A6 citing Li, Pearson-Merkowitz & Yokum 2026, arXiv:2606.14621v1), decide from the cited extract alone whether it is supported, not_supported, or not_assessable, with a locator and a one-sentence reason.
Data and exclusions
Inputs read: data/assertions.json (6 items), data/README.md (14 lines), data/sources/arxiv-2609.27732v1.txt (134 lines), data/sources/arxiv-2606.14621v1.txt (58 lines). Only the extracts were used as evidence; no outside knowledge of the papers was used. Nothing was excluded.
Method
out/analysis.py parses the relevant table rows (Table 6 and Table 7 of the DCR paper; Table 2 of the RESEA paper), prints the source lines it relies on, and applies the task's conventions: % change = 100·(exp(b) − 1); 95% CI = b ± 1.96·SE (2.0 also checked). An assertion counts as supported only if every element (sign, magnitude, uncertainty, significance, population) matches. For A5 the script searches the DCR extract for any mention of men/women/male/female/sex/gender and finds none.
Results
| id | verdict | locator | decisive element |
|---|---|---|---|
| A1 | supported | arxiv-2609.27732v1 | Table 6 | row "ATT" / "% change" | column "5 km" | The table prints ATT −0.431** (0.210) and −35.0%; ** means p<0.05; the text says the estimate is significant at 5%; 100·(exp(−0.431)−1) = −35.0%. |
| A2 | not_supported | arxiv-2609.27732v1 | Table 7 | row "HIV" | column "1 km" | The 1 km HIV ATT is +0.063 (SE 0.801), not −0.063. The sign is wrong, so the claim that HIV hospitalisations fall is contradicted. |
| A3 | not_supported | arxiv-2609.27732v1 | Table 6 | row "% change" | column "3 km" | −0.283 log points is printed as −24.7% (exp conversion gives −24.65%), not 28.3%. The assertion treats log points as percent. |
| A4 | not_supported | arxiv-2609.27732v1 | Table 7 | row "Hepatitis" | column "5 km" | ATT −0.406, SE 0.250, so the 95% CI is [−0.896, 0.084], which contains zero. The table gives no star. The asserted [−0.66, −0.16] is ±1 SE. |
| A5 | not_assessable | not reported in arxiv-2609.27732v1 | The extract reports estimates by buffer, outcome and estimator only. Nothing is broken down by sex. |
| A6 | supported | arxiv-2606.14621v1 | Table 2 | row "Weeks on UI" | column "Average treatment effect estimate" | ATE −1.99 weeks, 95% CI [−2.24, −1.75], control mean 17.3 weeks; 1.99/17.3 = 11.5%, which matches the Section 3 text. |
Counts: 2 supported, 3 not_supported, 1 not_assessable.
Uncertainty
- A1 depends on the significance threshold. Using the Table 6 SE (0.210), |t| = 2.05 > 1.96, and the table's ** flag directly states p<0.05. With the 2.0 critical value the result is borderline (|t| = 2.05 is still > 2.0). With the Table A19 SE for the same estimate (0.218), |t| = 1.98: this is significant at 1.96 but not at 2.0. A19 still flags the estimate **. The verdict rests on the printed stars and text, so it holds. The 35.0% figure is printed and also follows from the exp conversion.
- A3 depends on the log-to-percent convention. Under the prescribed exp conversion the figure is −24.65%, and the printed value is −24.7%. A naive reading of "log points ≈ percent" would give 28.3%, but the task and the source both rule that reading out. The verdict does not change under either exact convention (exp, or the printed value).
- A4 depends on the critical value. The CI with 1.96 is [−0.896, 0.084] and with 2.0 it is [−0.906, 0.094]. Both contain zero, so the verdict is robust. Only a ±1·SE interval (about 68% coverage) would reproduce the asserted bounds.
- A2, A5, A6 depend on no convention. A6's 11.5% is an exact ratio and is also printed in the text.
Limitations and open issues
- The DCR extract is internally inconsistent in several places. These are recorded in
data_issues, and none changes a verdict:- Table 6 gives the SE of the same CS-DR 5 km estimate as 0.210, while Table A19 gives 0.218.
- The A19 note says 640 control regions; everywhere else says 608.
- The abstract says fatal hospitalisations fall by 35%, but Table 7 implies −29.8% and Section 6.1 says about 30%.
- The prose calls the HIV estimates "close to zero" even though the 5 km estimate is −0.340.
- The RESEA text writes the weeks-on-UI CI as positive magnitudes, and there is a rendering defect ("p ¡ 0.001").
- A5 is
not_assessableonly with respect to the extract. The omitted parts of the paper might contain a breakdown by sex, but they are not admissible evidence here. - A3 is also incomplete in another way: it omits that the 3 km estimate is statistically insignificant (SE 0.279). That omission alone would not decide the verdict.