Tag
This paper introduces a claim-centric auditing framework for identifying error spans in deep-research agent trajectories, along with a new benchmark TELBench, improving process-level reliability assessment.