KEEPER: Learning Control-Sufficient Textual Memory for Long-Horizon LLM Agents

A new arXiv paper (2610.41482v1) introduces KEEPER, a comparative analysis of four auxiliary objectives for recursive textual memory in long-horizon LLM agents, evaluated along information-content and update-sensitivity axes within a shared recursive interface. The authors report that reward-related prediction achieved the highest task performance both before and after offline reinforcement learning, outperforming joint supervision, and define "self-consistency bias" as the persistence of incorrect claims already present in incoming memory. A claim-level analysis of one baseline trajectory documented error persistence despite corrective observations, leading the authors to argue that error persistence and valid-information retention should be evaluated alongside task outcomes.

arXiv:2610.41482v1 Announce Type: new Abstract: Recursive textual memory can carry earlier errors into later updates, yet task performance alone does not establish memory reliability. We define self-consistency bias as the persistence of incorrect claims already present in the incoming memory. We present KEEPER, a comparative analysis of four auxiliary objectives along information-content and update-sensitivity axes within a shared recursive interface, examining how supervision targets relate to agent performance. Among evaluated conditions, reward-related prediction achieve the highest task performance before and after offline reinforcement learning, outperforming joint supervision. A claim-level analysis of one baseline trajectory documents error persistence despite corrective observations. Together, these analyses motivate evaluating error persistence and valid-information retention alongside task outcomes.