Incident Postmortems
Postmortems are immutable historical records except for clearly marked corrections. A current audit of the index does not revalidate every external fact in an incident report.
Log
| Date | Incident | Severity | Audit state |
|---|---|---|---|
| 2026-03-14 | RunPod image regression | Critical | GPU-owned evidence excluded from the 2026-07-23 audit |
When one is required
- any critical incident;
- user-impacting data loss or a credible near miss;
- a security or privacy incident;
- a multi-user outage lasting more than 15 minutes;
- a partial/failed deployment that changes production behavior unexpectedly.
Process
- Mitigate first and preserve timestamps, revisions, IDs, and logs.
- Copy the template.
- Separate observed facts from hypotheses.
- Identify system conditions, not an individual to blame.
- Give every action item an owner, due date, and verifiable completion test.
- Have a second responder review it and add it to this table.
Related: Runbooks and Observability.
Was this page helpful?