Morning Edition · Wednesday, October 7, 2026Published at 2:04 AM EDT · New York

New papers question what reinforcement learning guidance, backdoor leaderboards and summary-tree retrieval actually measure

One paper gives a bias-variance theory of when external guidance helps reinforcement learning with verifiable rewards, another says backdoor-defense costs require a never-poisoned control model, and a third retests RAPTOR-style trees at matched cost.

Three of Tuesday's arXiv postings are about measurement more than new methods. Each argues that a common comparison in its subfield confounds two effects. The first paper addresses reasoning post-training. Reinforcement learning with verifi…