1# F4-conf-02 — Are the six quoted assertions supported by the sources they cite?23## Question4A literature review makes six assertions (A1–A6), each citing one of two arXiv extracts: Armendariz (2026), arXiv:2609.27732v1, on drug consumption rooms (DCRs), and Li, Pearson-Merkowitz & Yokum (2026), arXiv:2606.14621v1, on the Rhode Island RESEA programme. For each assertion I decide, using only the cited extract, whether it is **supported**, **not_supported** or **not_assessable**. Each verdict comes with a locator and a one-sentence reason.56## Data and exclusions7I read `data/assertions.json` (6 items), `data/README.md` (14 lines), `data/sources/arxiv-2609.27732v1.txt` (134 lines) and `data/sources/arxiv-2606.14621v1.txt` (58 lines). The SHA-256 hashes of both extracts match the ones listed in the README. Nothing was excluded, and I used no outside knowledge of either paper.89Problems found in the inputs (also listed in `results.json` → `data_issues`):10- **DCR, Table A19 vs Table 6:** the same CS-DR 5 km ATT (−0.431**) is printed with SE 0.218 in Table A19 and SE 0.210 in Table 6.11- **DCR, Table A19 note:** it says there are 640 comparison regions, but the table row and Tables 6–7 say 608. The note is also cut off mid-sentence at the end of the extract.12- **DCR, abstract vs Table 7:** the abstract says fatal hospitalisations fall by 35%. Table 7 (−0.354 at 5 km) implies −29.8%, and Section 6.1 says "approximately 30%".13- **DCR, HIV wording:** the text calls the HIV estimates "close to zero", although the 3 km and 5 km point estimates are −0.239 and −0.340 (both insignificant).14- **RESEA, Section 3 text:** confidence intervals for the weeks-on-UI reductions are written as unsigned magnitudes ([1.75, 2.24]). There is also an encoding defect: "p ¡ 0.001" stands for p < 0.001.1516None of these issues changes a verdict.1718## Method19For each assertion I located the decisive table cell and checked every element: sign, magnitude, uncertainty, significance and population.20- Log-point coefficients are converted to percentage changes as 100·(exp(b) − 1).21- 95% confidence intervals are computed as b ± 1.96·SE, with ±2.0·SE as a robustness check.22- Significance is read from the printed stars (** = p < 0.05).23- For A5, I searched the whole DCR extract for any breakdown by sex (men, women, male, female, sex, gender).2425The logic is coded in `out/analysis.py`, which prints the source lines it relies on and regenerates `out/results.json`.2627## Results28| id | verdict | locator | decisive element |29|---|---|---|---|30| A1 | supported | arxiv-2609.27732v1, Table 6, row "ATT", column "5 km" | The table prints −0.431** with SE (0.210) and a % change of −35.0%; 100·(exp(−0.431) − 1) = −35.0%; ** means p < 0.05. |31| A2 | not_supported | arxiv-2609.27732v1, Table 7, row "HIV", column "1 km" | The printed coefficient is **+0.063** (SE 0.801), not −0.063, so the claimed sign (a fall) is wrong. |32| A3 | not_supported | arxiv-2609.27732v1, Table 6, row "% change", column "3 km" | −0.283 log points is a −24.7% change (the table prints −24.7%), not −28.3%. The assertion treats log points as if they were percent. |33| A4 | not_supported | arxiv-2609.27732v1, Table 7, row "Hepatitis", column "5 km" | With ATT −0.406 and SE 0.250 (no star), the 95% CI is [−0.896, 0.084], which includes zero. The claimed [−0.66, −0.16] is only ±1 SE. |34| A5 | not_assessable | not reported in arxiv-2609.27732v1 | The extract has no estimates by sex: no search term matched any line. |35| A6 | supported | arxiv-2606.14621v1, Table 2, row "Weeks on UI", column "Average treatment effect estimate" | The table prints ATE −1.99 weeks, 95% CI [−2.24, −1.75] and a control mean of 17.3 weeks; 1.99/17.3 = 11.5%, which the text also states. |3637**Counts:** 2 supported, 3 not_supported, 1 not_assessable.3839## Uncertainty40No sampling or randomness is involved in the verdicts; the remaining uncertainty is about conventions:41- **A4 depends on the critical value.** The CI includes zero with 1.96·SE ([−0.896, 0.084]) and with 2.0·SE ([−0.906, 0.094]). Even a 90% CI (±1.645·SE, [−0.817, 0.005]) just includes zero, and the table prints no star. The claimed interval can only be reproduced as ±1·SE, which is roughly a 68% interval. The verdict does not change under any reasonable alternative.42- **A3 depends on the log-to-percent conversion.** Under the prescribed conversion the implied change is −24.65%, which the source itself prints as −24.7%. The figure −28.3% would only follow from reading log points directly as percent, which the task rules out, so the verdict does not change.43- **A1 uses both conventions.** The exp conversion gives −35.01%, consistent with the stated 35.0%. Significance at 5% comes from the printed stars, and it also holds by z = 0.431/0.210 = 2.05. Using the Table A19 SE of 0.218 instead gives z = 1.98, which is still above 1.96. Under the 2.0 critical value that alternative SE would sit just below the threshold, but the assertion quotes SE 0.210 and the ** of Table 6, so the verdict stands.44- **A6 involves only arithmetic** (11.50%) and quoting printed values.45- **A2 and A5 involve no convention.**4647## Limitations and open issues48- The verdicts check fidelity to the extract only, not whether the underlying studies are valid. For example, the DCR paper's own inconsistencies (abstract vs Table 7, SE 0.210 vs 0.218, 608 vs 640 controls) are recorded but not resolved.49- A5 is not_assessable from the **extract**. The full paper (for instance its omitted appendices) might contain a split by sex, but under the information rules that cannot be checked here.50- Locators point to the main tables. Where the text restates a value, it agrees with the table: A1 and A6 are confirmed in the text as well.51