postmortem-facilitation

Postmortem facilitation

Read composition guidance for optional expertise, fallback and scope.

Reconstruct how the system and response produced the observed impact. Separate the event that triggered the incident from the conditions that let it propagate, the signals that revealed it, and the actions that restored service. Preserve what responders knew at the time so hindsight does not turn a reasonable decision under uncertainty into an unsupported account of individual failure.

Use each question to resolve a gap in that explanation. Ask for the observation or decision context that would distinguish competing accounts, then carry the answer into the causal analysis and prevention plan. A useful review can establish recovery while leaving part of the underlying cause unknown.

Modes

• Interactive interview (only when requested): review provided history first, maintain an evidence ledger, ask one substantive question at a time and follow its answer. Summarize newly established facts and explain why the next gap matters.
• Draft/review: assemble what sources establish, challenge unsupported claims and return targeted unresolved questions. Do not require an interview when the evidence already answers them.
• Recurring review: identify incidents needing review, assemble drafts/question queues, detect contradictory or weak causal accounts and prepare grounded action candidates. Do not invent participant answers, endorse ownership on their behalf or publish a draft as approved.

Consult references/facilitation-guide.md for deeper implementation details when needed.

Facilitation workflow

1. Ingest ticket, chat/timeline, changes, metrics/traces/logs, actions and stakeholder communications. Normalize timestamps with source/timezone and separate customer impact from response and ticket duration.
2. Create a fact/claim/unknown ledger and a first causal/response timeline. Read system evidence before asking questions answerable from code or history.
3. Ask adaptively about the highest-value gap: actual user harm, failure mechanism, decision context, detection gap, mitigation delay, recovery evidence or recurrence protection.
4. Challenge contradictions neutrally: cite competing evidence, ask what reconciles it, retain both until resolved. Use “what information/control made this outcome possible?” rather than “who failed?”
5. Analyze trigger, contributing conditions, propagation, barriers, detection, response and recovery. Do not force all incidents into one root cause or a fixed five-whys chain.
6. Derive specific learning and candidate actions with a causal rationale and observable completion criterion. Distinguish confirmed correction from hypotheses to investigate.
7. Draft the review in the local format with impact, timeline, analysis, what worked, gaps, unknowns and actions. Mark owner/reviewer confirmation pending where needed.
8. Review for factual accuracy, psychological safety, useful depth and testable prevention. Hand approved work to postmortem-followup or return the register directly.