发现智能体在推理前就因未读检索结果而失败，提出强制阅读机制提升准确率。
Before Reasoning Can Fail: Pre-Evidence Procedural Failures in Agentic RAG
- 通过分析工具调用轨迹，区分检索前与读后失败两类错误模式。
- 强制阅读后，模型准确率提升3.2至19.9个百分点，尤其对跳过阅读的路径效果显著。
- 提示大思考预算不等于更好证据利用，需独立评估证据获取过程。
智能体检索增强生成系统可能在证据条件推理之前就失败：智能体虽检索到候选片段，却未检查即得出结论。我们将其失败模式视为智能体轨迹的程序性问题，利用保存的工具调用记录、检索证据、阅读段落和最终答案，将错误分解为检索前纪律缺失和读后黄金读取失败。在HotpotQA、2WikiMultiHopQA和MuSiQue上共12,000对轨迹中，两类失败基本非冗余，双触发率在[11.2%, 13.1%]之间（使用正则表达式和spaCy实体提取器）。随后评估Read-Gate，一种最小运行时不变量，要求智能体在搜索后、最终输出前必须阅读。强制阅读使原本跳过阅读的轨迹中LLM-Acc提升14.9-19.9点，全最小推理单元中提升3.2-9.4点。额外诊断显示，更大的隐式思考预算并不一定提升证据检查率。结果表明，证据收集应作为轨迹级控制问题单独评估，而非仅关注答案推理。
原文摘要 · Abstract (English)
Agentic retrieval-augmented generation (RAG) systems can fail before evidence-conditioned reasoning is tested: an agent may retrieve candidate snippets but finalize without inspecting them. We study this failure mode as a procedural property of the agent trajectory, decomposing wrong answers into pre-evidence discipline failures and post-gold-read failures using saved tool-call traces, retrieved evidence, read passages, and final answers. Across 12,000 paired trajectories on HotpotQA, 2WikiMultiHopQA, and MuSiQue, the two failure types are largely non-redundant: the both-trigger rate is in [11.2%, 13.1%] across regex and spaCy entity extractors. We then evaluate Read-Gate, a minimal runtime invariant requiring an agent to read after search and before finalization. Forced reading improves LLM-Acc by 14.9-19.9 points on trajectories that would otherwise skip reading and by 3.2-9.4 points on full minimal-reasoning cells. Additional diagnostics show that larger hidden thinking budgets do not necessarily increase evidence inspection. Together, these results indicate that evidence-gathering should be evaluated as a trajectory-level control problem, separately from answer-side reasoning.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。