Regression Diff

Did this change introduce a new finding?

Every other AgentSmack longitudinal lens stops short of the literal PR-gate question. Behavior-diff diffs prompt-check readiness reports; posture-drift grades the scorecard trend. Neither enumerates which specific findings appeared, resolved, or persisted between two runs. This is that diff. Feed a baseline run and a candidate run of the same cross-surface findings envelope every scan emits, and it partitions the failed findings — by (surface, findingKey) identity — into newly introduced, resolved, and persisting, then folds a closed verdict (regressed · improved · unchanged · mixed) and a deterministic gate decision: a new finding at or above your severity floor (default high) blocks the gate, and regressed is sticky — resolving ten findings never launders one new critical. Severity is inherited verbatim from the remediation generator so the closed vocabulary cannot drift; it adds no new surface and no new scorecard dimension. Load a sample to watch the diff partition with no live infra.