# Reproduction and hazard notes

Reviewer model: gpt-6 (Codex). This is a computational reproduction, not an endorsement of causal or clinical interpretation.

Read all bundled code, claims, paper, plans, deviations, requirements, materials, provenance, references and results. Both registered Python files are byte-identical to their executed counterparts. The post hoc classifier calibration is disclosed. No prompt instructions aimed at the verifier were identified.

The sole node integrity flag is the absent Discussion heading. Interpretation is present in Results and Limitations, but the formal heading is missing. This formatting issue does not prevent reproduction. A wording concern beyond reproduction: the bootstrap CI itself is not Holm-adjusted; the code adjusts the two primary p-values. The paper's claim-section description should keep this distinction clear. Non-significant diabetes or coherence tests are not evidence of equivalence.

Hazard assessment: none. The paper states holder-deidentified public release, and the source is an open-access PhysioNet database with a CC-BY license and IRB information: https://physionet.org/content/cerebral-vasoreg-diabetes/1.0.0/ . This is analysis of released physiological measurements, not a capability for mass harm or unauthorized cyber activity. No credentials or new personal identifiers were found. The existing source subject codes remain deidentified and are not linked to people.

The reference harness obtains and hashes all 63 external files, builds the pinned environment and runs the feature extraction and analysis with container isolation and no runtime network. Its machine-readable comparison is the basis of the reproduction verdict. An independently written statistics script additionally recomputes the main counts, Hodges-Lehmann estimate, Mann-Whitney AUC, signed-rank tests and Holm correction from the newly generated feature table. It does not import submitted code; it uses this review environment's SciPy 1.15.3, so tiny numerical differences should be evaluated against declared tolerances.