# Contract — F4-conf-04

Question. A draft cites five arXiv studies (R1–R5). For each one, check against the provided extract only: (i) does the entry title match the document title? Minor typographic differences are allowed. (ii) Is the entry year equal to the first-submission year or to the year of the reproduced version? (iii) Is every element of "cited_for" (magnitudes, SEs/CIs, significance, timing, population, design) stated in the extract or arithmetically implied by it? Estimand. The set of unverifiable references (those failing at least one check), plus a per-reference record {title_matches, year_matches, finding_supported, locator, note}. Assumption: the extracts faithfully reproduce the documents. The headers say "no numbers were altered". Outputs. out/results.json with keys reference_checks, unverifiable_references, n_verified, task_id="F4-conf-04", feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed=null. Also out/report.md (Question; Data and exclusions; Method; Results; Uncertainty; Limitations and open issues), out/analysis.py (deterministic, reads data/ only, prints the header lines and table cells it relies on), and out/claims.json. Information rules. Only extract headers and bodies are evidence. No memory of the papers and no external databases. Report numbers as printed. Invalidity conditions. A missing or unreadable extract, or a missing Bibliographic record line, would make a check infeasible. Using outside knowledge would invalidate the result.