Regression Triage¶
Use sentinel-dv-regression-triage when a run failed, pass rate changed, or a regression needs an evidence-backed priority order.
Tool sequence¶
runs.list
-> runs.get + runs.summary
-> regression.health
-> regressions.summary (suite history, when needed)
-> tests.cluster
-> runs.diff (when a baseline exists)
-> failures.list + assertions.failures
-> tests.history / runs.cross_sim
-> targeted tests.get / topology / wave tools
The skill does not call every tool mechanically. Compile and elaboration failures stop before waveform analysis; runtime failures can continue into topology and bounded waveform windows.
Example request¶
Use the Sentinel DV regression triage skill on run R124.
Compare it with R123, rank signature clusters, and disclose any unavailable
health components or truncated pages.
Report contract¶
The result includes scope and counts, health data quality, baseline changes, prioritized clusters, heuristic limits, actions, and missing evidence. A cluster is a signature grouping, not proof of a shared root cause.
Interpreting health¶
regression.health is a scoped indicator. Components without sufficient source data are null and removed from effective_weights:
- coverage needs indexed metrics;
- assertion health needs indexed SVA status;
- flakiness needs repeated test cohorts;
- cross-simulator consistency needs comparable cohorts from at least two simulators.
Do not use the score as a sign-off gate without reviewing data_quality.