Operational health: deploys, crons, alerts, anomalies.
How long failed scheduled jobs take to recover.
Operational alert volume over time, by severity.
Rule-based detection over the unified audit log.
Commits merged to main as a deploy-frequency proxy.