Within 24 hours of an error, write one mechanistic sentence + the incorrect assumption it revealed — before memory reconstructs

Why This Is a Rule

Memory of errors degrades in two predictable ways within 24-48 hours. First, the mechanistic details blur: you remember that something went wrong but not the specific sequence of events that produced the failure. Second, hindsight bias reconstructs the memory to make the error seem more predictable (Review decisions in three steps: re-read reasoning blind, predict outcome, then compare — this sequence defeats hindsight bias) — "I should have known" replaces "I couldn't have known," which prevents accurate assumption identification.

The one mechanistic sentence captures the causal sequence while it's still sharp: "I deployed without checking the staging environment because I assumed the staging and production configs were identical." This sentence contains the mechanism (deployed without checking) and, implicitly, the incorrect assumption (configs were identical). Within 24 hours, you could reconstruct this accurately. After a week, it becomes "I made a deployment mistake" — which carries no learning.

The assumption identification is the learning payload. Every error reveals at least one assumption that was wrong. Identifying it converts the error from a negative event into a falsified hypothesis — which is information you can use to prevent the category of error, not just the specific instance.

When This Fires

• Within 24 hours of any error, mistake, or unexpected negative outcome
• Before the emotional charge of the error fades (which also lets the mechanistic memory fade)
• As the quick-capture complement to Log every mistake for 30 days with date, event, and conditions — no analysis, just raw data for pattern detection's 30-day error log (this provides the per-error detail)
• When you want to extract the maximum learning from each error without waiting for aggregate analysis

Common Failure Mode

Waiting more than 24 hours: "I'll do a proper retrospective next week." By next week, the mechanistic details have blurred and hindsight bias has reshaped the memory. The retrospective produces a reconstructed narrative that confirms what you already believe rather than revealing the actual failed assumption. The 24-hour window is not flexible — it's the deadline before memory corruption makes accurate capture impossible.

The Protocol

(1) Within 24 hours of an error, write one sentence describing the mechanism: what happened, mechanistically, stripped of emotion and judgment. Not "I screwed up the deployment" but "Code was deployed to production with a configuration value that differed from staging because the values were manually maintained in two files." (2) Then identify the single incorrect assumption: "I assumed the staging and production configuration files were synchronized." (3) The assumption should be something you believed was true that turned out to be false. It's the gap between your mental model and reality that the error exposed. (4) Store both the mechanism and the assumption. Over time (Log every mistake for 30 days with date, event, and conditions — no analysis, just raw data for pattern detection, After 10+ post-action reviews, analyze in aggregate — patterns across unrelated tasks reveal systemic tendencies, not isolated errors), your collection of falsified assumptions reveals your systemic blind spots — the categories of things you consistently get wrong.