Results for “reaction-mechanisms”

Detect poisoned training data and backdoored models across the ML pipeline using statistical analysis, activation clustering, and spectral signatures.

`analysis-agent`/`task-agent`/`review-agent`: use for release identity, compatibility, exposure, stop, rollback, or recovery; skip pipeline-only and no-release work.