Where Do Deep-Research Agents Go Wrong? Span-Level Error Localization in Agent Trajectories
2026-06-01 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors study how deep-research agents make mistakes during their step-by-step problem-solving process. Instead of only checking if the final answer is right or wrong, they look closely at which parts of the agent's reasoning went wrong. They collected many real examples, labeled error spots, and created a benchmark called TELBench to help identify these errors. They also introduced DRIFT, a method that tracks the agent's claims and finds where unsupported or conflicting information leads to mistakes. Their approach improves the detection of errors within the reasoning process, giving a clearer view of agent reliability.
deep-research agentstrajectoryerror localizationTELBenchDRIFTclaim auditingLLM-assisted reviewspan-level annotationbenchmarkanswer synthesis
Authors
Jiaming Wang, Ziteng Feng, Jiangtao Wu, Ruihao Li, Qianqian Xie, Yuxiang Ren, He Zhu, Xueming Han, Fanyu Meng, Junlan Feng, Jiaheng Liu
Abstract
Deep-research agents solve tasks through long trajectories of search, tool use, evidence inspection, and answer synthesis. Evaluation based on final answers shows whether an agent succeeds, but not which parts of the trajectory make the answer unreliable. We study span-level error localization for deep-research agents. We collect 2,790 real trajectories from two agent frameworks, three backbone models, and three benchmarks, convert raw logs into semantic spans, and annotate harmful error spans through LLM-assisted expert review. From these annotations, we build TELBench, a 1,000-instance benchmark for identifying error spans among normal exploration, failed searches, tentative hypotheses, and harmless noise. We further propose DRIFT, a claim-centric auditing framework that tracks agent claims, checks their support in trajectory evidence, and marks spans where unsupported or conflicting claims affect the answer path. Experiments across model families and auditing frameworks show that DRIFT improves span-level error localization and first-error accuracy by up to 30 percentage points. Our work provides a process-level view of reliability in deep-research agents.