Interlock: From Local Correctness to Task Completion in Long-Horizon Agents.

arXiv:2610.19686v1 Announce Type: cross Abstract: Large language model agents increasingly undertake long-horizon tasks in complex environments characterized by fragmented evidence, partially completed states, and dependencies across resources and artifacts. E