audit-training-experiment

Audits an ML training experiment for reproducibility, evidence integrity, leakage, checkpoint identity, evaluation validity, and overclaimed conclusions. Use before accepting results, spending on a longer run, publishing artifacts, or handing work to another agent.