Operator runbook index
Symptom-first routing to the owning operational contract, evidence and stop condition.
First five minutes
- ✓Confirm environment, release and whether a deployment/change is in progress
- ✓Check /health, /ready and /health/clock without restarting first
- ✓Open active/failed OperationRuns and identify the owning step
- ✓Capture error_code, correlation and dependency/provider identifiers
- ✓Freeze duplicate submissions and unrelated mutations
Symptom routing
| Symptom | Go to | Stop condition |
|---|---|---|
| API not ready | Observability & support | Do not enable lifecycle flags |
| Agent missing/incompatible | Install the agent | No work assignment |
| Provider validation/throttling | Cloud providers | No client retry loop |
| Cluster run stuck | OperationRuns | Do not create a duplicate intent |
| Add-on health failed | Built-in add-ons | Do not open dashboard or install dependants |
| Evidence checksum failed | Troubleshooting | Treat artifact as unusable |
| Backup verify failed | Backup & restore | Block upgrades |
| Unknown stable code | Error catalog | Escalate if absent |
During a failed change
Use the OperationRun as the incident timeline. A service restart is not a universal recovery action: it can move leases, obscure in-memory clues or trigger startup recovery. Decide cancel, retry, cleanup, rollback or restore from the durable state contract.