Agentic Interpretation：harness 治理的软件智能体证据构建
原标题：Agentic Interpretation: Harness-Governed Evidence Construction for Software Agents
AI 摘要
论文提出 Agentic Interpretation (AgInt) 框架，让模型提出类型化记录，由确定性 harness 负责准入、评分与持久化，并仅在证书满足任务义务时放行答案。在 SWE-bench-Live、DeepSWE、Defects4J 和 GitBug-Java 上，AgInt 在补丁接受度指标上优于普通 agent 与 Agentic Code Reasoning，并在故障定位中排名更准、检查无关行更少。以 GLM-5 为例，SWE-bench-Live 的 reject 类 F1 与 Defects4J 的 All@1 均有提升，GitBug-Java 的无效检查行数下降；同时执行次数更少、单候选成本更低。
正文节选
Agentic Interpretation: Harness-Governed Evidence Construction for Software Agents Abstract. Software agents answer behavioral questions about code, such as whether a patch will pass its tests or which lines caused an observed failure. These tasks require gathering and combining evidence across a repository, yet most agent scaffolds leave intermediate findings in a free-form trajectory, making them difficult to reuse and easy to disconnect from the final answer. We present Agentic Interpretation