Postmortem Portfolio Prioritization Prompt

You are a reliability lead deciding where a small amount of engineering time should go next quarter, based on a stack of recent postmortems. You resist recency bias — the last big outage is not automatically the right thing to fix.

I will paste:

[POSTMORTEM SET: summaries of N recent incidents, each with severity, duration, affected systems, and named contributing factors and action items]
[CONSTRAINTS: roughly how much reliability time is available, and any systems off-limits this quarter]

Do the following:

1. Extract a normalized list of contributing factors and proposed action items across all incidents. Cluster items that are really the same underlying gap described in different words.
2. For each cluster, estimate leverage: how many incidents it touched, the combined severity/duration it contributed to, and whether it's a prevent / detect / mitigate gap.
3. Rank clusters by leverage-per-effort, using effort as a coarse S/M/L since I haven't given you estimates — mark every effort guess as [ESTIMATE].
4. Call out the recency trap explicitly: which highly-visible recent item ranks lower than it feels like it should, and which quiet recurring gap ranks higher.
5. Recommend a short "fund these N" shortlist and name what you'd consciously not fund this quarter.

Output format: a clustered factor table with incident-count and leverage, then a ranked shortlist, then a short "what we're choosing not to do and why."

Guardrails: stay blameless — cluster on systems and gaps, never on which team or person caused more incidents. Mark every effort and leverage estimate as [ESTIMATE] until I supply real numbers. This informs prioritization; the human reliability owner makes the call.

Why this prompt works

Most reliability investment is decided one incident at a time, which guarantees the wrong answer. The team fixes whatever just hurt the most, the action items from that one postmortem get prioritized, and the quiet gap that contributed to six smaller incidents over the quarter never surfaces because no single one was loud enough to demand attention. Prioritizing across the whole portfolio of postmortems — rather than incident by incident — is how you find the gap that’s worth more than the last fire.

This prompt does the cross-incident normalization that nobody has time to do by hand: it pulls contributing factors and action items out of a stack of postmortems and clusters the ones that are the same underlying problem wearing different words. “Config drift,” “the deploy didn’t validate the manifest,” and “staging didn’t match prod” might be three descriptions of one gap, and only seeing them together tells you it’s worth real investment. Ranking by leverage-per-effort, with incident-count as a signal, is exactly the kind of synthesis that turns a pile of documents into a decision.

The recency-trap step is the part that earns its keep. By explicitly naming which visible recent item ranks lower than it feels like it should — and which boring recurring gap ranks higher — the prompt counteracts the bias that drives most reliability roadmaps. The guardrails keep two failure modes out: every effort and leverage number is marked as an estimate so a shortlist isn’t mistaken for a funded plan, and clustering stays on systemic gaps rather than ranking teams by incident count, which is just blame at scale. The human reliability owner still makes the call; the model just makes sure the call is informed by all the data, not the loudest slice of it.