Short accounts of decisions made while building: a rule I needed to reconsider, instructions I could no longer trust, or a correct result the interface did not make clear.
A piano accompaniment sounded too heavy, and I could not point to the notes. Finding them through conversation and listening, changing only those, and discovering that the score had already said something similar.
I designed my exercise log never to judge. Seven weeks into a plan with two strength sessions recorded, that same design keeps it from stating the gap.
A proactive AI feature worked before later changes exposed gaps in its integration. I trace what product requirements, shared boundaries and tests need to preserve.
Consolidating Battuta’s operation lists reduced duplication but left the command help free to drift. I used the parser’s own rejection to check what the help promised.
I had called MIDI’s default time signature a guess. Comparing the same notes under stated and default meter separates my editor’s policy from a format rule.