Inducing Reward-Free Judging Rubrics that Reduce Over-Crediting in Agent Evaluation

Researchers explore inducing reward-free judging rubrics to reduce over-crediting in language-model agent evaluations, addressing limitations of existing…