# Verifier notes

- Re-ran code/run in the bundle's pinned Python image with no network: exit 0 in about 9 s; every declared R1 result for C1 agreed within tolerance.
- Source: data/source.RDS is byte-identical to Noetel et al.'s OSF file (https://osf.io/download/68xnt/, SHA-256 50584b02...7781, downloaded today). Converting it with base R 4.4.2 (rocker/r-ver, no network) and write.csv as code/convert_source.R does gives the same 828 rows and columns; the only differences are 19 cells whose 15th significant digit prints differently (R 4.4 vs the stated 4.6.1), none in a field the analysis selects on.
- Hidden-content scan: the six control characters (U+0008, U+0001, U+0005) in data/source.csv lines 53-54 are inside a free-text notes field copied from the source and are also present in my own conversion of the original RDS. They sit where "r =", "=" and a minus sign would be ("(\b0.28; p \x01 .009)", "\x054.8 points"), the usual residue of PDF text extraction. They carry no instructions and do not steer a verifier.
- Benford flags (length, pre_n, n, mean, mean_diff, smd): these are trial durations, sample sizes, and scale scores in narrow ranges (weeks, tens of participants, questionnaire points), which are not expected to follow Benford's law; the data come unchanged from the published OSF extraction.
- Independent check (evidence/independent-check.txt), separate code in a python:3.12-slim container with numpy and scipy: recomputed each trial's Hedges g with the exact gamma correction from the arm summaries in study_effects.csv (max difference from declared 2e-14), then REML by direct bounded minimization of the restricted likelihood and the modified Hartung-Knapp interval. k = 21, tau^2 = 0.7370, g = -0.762, CI [-1.203, -0.321], matching the declaration. The prediction interval matches the declared [-2.613, 1.088] when the pooled-estimate variance inside it is the Hartung-Knapp one; with the plain REML variance it would be [-2.611, 1.087]. Either way it includes zero, so C1 is unaffected. Reviewers may want Methods to say which variance it uses.
- Hazard screen: a meta-analysis of published exercise trials; none.