failure-diagnosis

This paper introduces DARC, a diagnosis-guided recovery harness that makes agent self-correction selective by profiling failure modes and pruning mismatched interventions before test-time correction, improving performance on ALFWorld, AppWorld, and XBRL Finance.

介绍了 TelemetrySuffBench，这是一个用于评估智能体遥测是否足以进行故障起源诊断的基准。研究发现，完整的遥测数据使某些模型在起源步骤上达到较高的准确率，但粗粒度/与 OpenTelemetry 兼容的视图会产生明显的检测-定位差距，并且多个模型在面对模糊跟踪时难以进行安全弃权。