Start with the request ID
- Check
/api/healthfor process liveness and/api/readyfor dependency readiness. - Capture the
X-Request-Idfrom the failing response. - Search structured logs for that ID and identify the first failing boundary.
- Inspect the audit history before retrying a mutating operation.
Failure patterns
The API will not start
Check configuration parsing, database path permissions, migration status, and whether the configured port is already in use.
The console cannot authenticate
Verify the API origin, cookie/session settings, CORS configuration, clock skew, and that browser code is not reading server secrets.
A provider call fails
Check the selected model, provider endpoint, secret injection, timeout, outbound policy, and provider response logs. Redact credentials before sharing diagnostics.
A release is blocked
Read the validation and evaluation result, approval requirement, policy decision, and current release state. Do not bypass a gate by editing persistence directly.
The website link is broken
Check the generated route under the configured base path. Internal links must use Astro’s base URL rather than an origin-root path.
Recover safely
If user impact is ongoing, route traffic back to the last verified content hash, preserve evidence, and open an incident record. Restore the database and CAS together; restoring only one can break manifest identity and audit interpretation.