A thorough postmortemisn't a closed incident.

Two things: how complete the postmortem document is, and whether the remediation is actually finished. It scores a blameless writeup across six sections — timeline, root cause, impact, detection & response, action items (owned and dated), and blameless framing — into a 0–100 completeness score, and separately counts the critical action items still open. The verdict is CLEARED TO CLOSE, FINISH ACTIONS, or NOT CLOSEABLE. It grades the document and the remediation state, never the people in the incident.

Because a complete writeup isn't a finished fix. The open-critical-action gate is dispositive: if even one critical action item is still open, the incident is NOT CLOSEABLE no matter how thorough the document is — because the thing that actually prevents a repeat hasn't shipped yet. In the worked example, the prompt-injection postmortem scores a perfect 100 but has one critical action still open, so it can't close. A clean close has to be earned on both completeness ≥ 80 and zero open critical actions.

The preventive work that stops the same class of incident from recurring — a missing guardrail, an unbuilt alert, a rollback path that wasn't there. You mark which action items are critical; the gate only counts those. Non-critical follow-ups (a doc update, a nice-to-have) are tracked but don't block the close. The point is to separate "we wrote it up" from "we shipped the fix that matters."

Blameless framing is one of the six scored sections, and the whole rubric rewards systems-and-process language over individual fault. It grades the document and the remediation state, not the people in the incident — there is no per-person score or assessment anywhere in it. That's deliberate: people only report what really happened when the review isn't a search for someone to blame, so the scoring is built to reinforce that.

No. It doesn't page, detect, or manage live incidents — it's the close-out discipline that runs after the firefight, when you're deciding whether the incident is actually done. It pairs with the upstream tools: the Agent Reliability Harness catches failures before they become incidents, and the AI Output Audit-Trail Kit gives you the record-keeping that makes a postmortem reconstructable. This is the gate at the end that refuses to call it closed early.

One .xlsx scoring workbook (the six-section completeness score + the open-critical gate + the verdict), a Facilitator Playbook, a Postmortem Template & Close-Out Runbook, and a worked six-incident example. It's deterministic and offline — opens in Excel, Google Sheets, or Numbers. It's a working aid for incident close-out discipline, not legal advice; some incidents carry regulatory or contractual reporting duties, so confirm those separately with the appropriate owner.