ww-mistake-ledger

The problem: The same mistake gets made twice, sometimes three times, each with the same surprised reaction, because it got written down once and never looked at again. A mistakes doc that’s only ever appended to isn’t a safety net, it’s a diary nobody rereads, and the note sits there correct and unread while the same shape of mistake happens again on a different piece of work.

This closes the loop on both ends: name the mistake plainly, root-cause the general class it belongs to rather than the one instance, fix it at the enforcement layer in the same sitting (a checklist item you’ll genuinely follow, a lint rule, a hook, not “I’ll remember”), and check the log before starting work that resembles a past entry, not from memory. A second occurrence of the same class is treated as proof the fix was too weak and gets escalated to something stronger, not logged again as a bigger entry.

Key principles

• Prevention is the finish line, not documentation; a written mistake that recurs unchanged hasn’t been closed
• Generalize one level up; the specific instance rarely repeats, the class almost always does
• A second occurrence is data, not bad luck: it means the fix needs to move up the enforcement ladder
• Reserve it for real, avoidable mistakes with a real consequence, not routine iteration
• Cheap and constant beats rare and heavy; a short entry checked before the next similar task beats an occasional deep audit

Part of Operator Skills, a collection of skills for running a real Claude Code setup.