# Verifier notes

- Re-ran `sh code/run` with the harness in Docker, built from the bundle's pinned `env/Dockerfile` (python:3.13-slim by digest; numpy, scipy, matplotlib pinned), with no network. Exit 0 after about 2 s.
- All 11 declared results for C1 agree within tolerance (exact to the declared precision).
- The run also regenerates `results/validation.json`, the comparison of the Python meta-analysis against the archived R/metafor reference output in `data/metafor-reference.json`; I did not rebuild the optional R environment.
- Hazard screen: none. The data files hold aggregate effect estimates, counts and short paraphrases from published cohort reports, with sources listed; no source text, abstracts or figures are reproduced beyond short factual statements, no patient-level or private data, no credentials, and no instructions aimed at verifiers were found.
