Container Registry and Software Supply Chain Platform · View 26 of 33 · 5 · Operations
First-class metrics
- Admission decisions by verdict and reason; policy evaluation latency
- Scan queue depth and age; advisory feed ingestion lag; signature verification failures
- Count of running workloads in violation of current policy — the number that says whether the platform is working
Decisions
- The audit trail is separate from the metrics pipeline and lands in an append-only ledger independent of the registry
- 100% of production admissions have an audit record; a missing record is a P1 incident
- A per-artifact evidence view shows every attestation, its signer, verification result and age
Alert discipline
- Scan backlog alerts rather than silently allowing; feed staleness becomes a policy input rather than an alert nobody acts on
- Exception ratio above threshold alerts, because a platform running mostly on exceptions is not a control
- Break-glass use always alerts, and the session is time-boxed