Contract (P0)
Question: extract, verbatim, specified treatment-effect estimates and their uncertainty from two arXiv extracts (Armendariz 2026 DCR staggered DiD, Tables 6/7; Li et al. 2026 RESEA RCT, Table 2), attach locators, derive (a) a 95% CI from the printed SE of the Table 6 5 km ATT and (b) an SE from the printed 95% CI of the Table 2 wage ATE, and identify the requested quantity that is not printed (exact p-value of HIV 5 km ATT).
Estimands: the printed numbers themselves (no re-estimation). Derived quantities are meaningful under an approximate normal sampling distribution of the estimator (CI = est ± 1.96·SE; SE = (hi − lo)/(2·1.96)); the RESEA CI is described as from "the methodology described in the main text" (not in extract), so symmetric-normal is an assumption.
Outputs: out/results.json (keys listed in TASK.md: dcr_* , ri_*, locators, derivation_notes, task_id, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed), out/report.md (6 sections), out/analysis.py, out/claims.json, plus protocol state files and out/exclusions.csv.
Information rules: only the two extracts are evidence; values as printed (sign, decimals) from the table named in the key; derived keys from the named table with stated convention; non-printed quantity → null with "not reported in
Invalidity: using values from another table (e.g. Table A19 SE 0.218 or Table 3), rounding/rescaling printed values, inventing a p-value, or using memory of the papers.