10:27
官方账号arXiv cs.AI@Srimonti Dutta, Akshata Kishore Moharir
推荐理由：Read this paper if you're interested in improving the reliability of LLM data agents and understanding the importance of Trace Integrity in real-world systems.
evaluation signals
共 1 条相关 AI 资讯
8月27日
10:27
This paper introduces Trace Integrity, a deployment reliability criterion for evaluating LLM data agents. It identifies the Structure Gap as the deployment failure mode and operationalizes Trace Integrity with execution contracts. The study shows that answer accuracy, trace validity, and silent-failure risk are distinct evaluation signals. Real-world LLM data agents should be evaluated by their auditable computation.