The outage is over. The graph is green again. Someone drops a postmortem link in Slack with five action items and a timeline that took two hours to write.
You open it. You nod. You react with 👀. Two months later the same class of failure shows up, and nobody can find the sentence that said "always set a timeout on that client." The doc still exists. The lesson does not.
That is the postmortem problem. Teams get better at writing them than at keeping the useful bits warm in people's heads.
Why the PDF and the wiki page both fail
A good postmortem is long on purpose. Timeline, impact, root cause, contributing factors, what went well, what did not. That length is right for the incident review meeting. It is wrong for the Tuesday six weeks later when you are about to ship a similar retry loop.
People treat the link as memory. Links are storage. Memory is the one or two constraints you can still say out loud when someone asks "what did we learn last time?"
Reading the full write-up again rarely happens. It feels like homework after a fire. So the action items rot in Jira, the doc sits in Confluence, and the next on-call inherits vibes instead of rules.
Pull out the rule, not the novel
Right after the review, while the argument is still sharp, capture the postmortem the same way you would save anything else you mean to keep. Browser save, share sheet, or Ctrl/Cmd+Shift+S if you use a Chrome extension. One library beats three Slack bookmarks and a Notion page you will never search under pressure.
What you need from that page is small. A short TL;DR. A handful of key takeaways. The one rule that would have prevented the repeat. Write that rule in your own words on a note card: "outbound HTTP to payments must have a 3s timeout and circuit breaker" beats a polished paragraph about "resilience."
I park those extractions in Recl. Share the postmortem URL or a screenshot of the action list, get a short summary and takeaways, drop a one-line "your note" while the review is fresh, and let the save land under a topic instead of a flat "incidents" dump.
Skip re-reading the full timeline in that first window. You just lived it. You need the portable rule, not another hour inside the document.
Put the lesson on a schedule you will actually open
Saving is not remembering. Incident knowledge fades between on-call rotations. You ship for a month, then blank on why that queue had a poison message policy.
Passive re-reading of the same Confluence page feels responsible and still leaves you stuck when a similar alert fires at 2 a.m. What helps is retrieval: close the doc and ask yourself what the timeout was, which dashboard lied, and which runbook step was missing. Wrong answers mark the fuzzy edge.
A light loop works. A few days later, a nudge asks whether you still remember this one. Answer from memory. If you blank, open the takeaway card, not the twenty-page write-up. On Sunday evening, a short weekly quiz over what you saved that week keeps a few rules warm without turning memory into a second job.
That is the Recall side of Recl. Nudges stay short on purpose. You are keeping a handful of incident rules warm enough that they show up before the next similar deploy, not memorizing every paragraph of every postmortem.
Any system that captures the rule and schedules a revisit works. Spaced repetition fits timeouts and ownership the same way it fits flashcards. The point is a reason to see the lesson again before the next outage asks for it.
Ask your library when the pager is loud
The real test is not the review meeting. It is a half-remembered incident from last quarter while a new alert is screaming.
Instead of scrubbing Slack for the old thread, ask your library. In Recl that means asking a question and getting an answer from your own saves, with sources, so you can jump straight to the takeaway that named the missing timeout. You pull from material you already chose to keep.
If you want to see how a clean takeaway card reads, browse library.recl.app. Your own postmortems still matter more. The point is a version you will open when the graph is red, not when you have an afternoon to re-read.
Folders and auto topics help here. Keep "payments timeouts" separate from "deploy rollbacks" so you are not scrolling a graveyard of every SEV-2 from last year.
A loop you can run after every review
- Capture the postmortem URL or the action list the same day as the review.
- Write one rule in your own words. If you cannot name a rule, the write-up did not finish its job.
- Keep the takeaways short enough to open under stress.
- Revisit on a light schedule. Fail the quiz honestly. Re-read your sentence, not the whole timeline.
- Before the next similar change, ask your library first instead of hoping Slack search remembers.
You do not need every postmortem you have ever starred. You need three rules from the last few months that you can still teach a new teammate at 1 a.m.
Postmortems fail as learning tools when the document becomes the product. Make the portable rule the product. Save it once. Bring it back on purpose. The next outage will still hurt. It does not have to hurt the same way twice.
Asutosh builds Recl — save what matters, recall it forever. Summaries, library topics, and spaced recall for the docs and videos you'd otherwise forget.
