PROJECT FOUR - EVALUATION AND OBSERVABILITY

EVALUATION-DRIVEN IMPROVEMENT The habit to build is a loop. Change something, run the whole eval set, look at the pass rate, and keep the change only if nothing else got worse. Never call a fix done because the one case you were staring at now works. Run all fifteen, because a fix to one place can quietly break another. Add a new eval item for every real failure you meet, so the same failure can never return unnoticed.