Search papers, labs, and topics across Lattice.
This study investigates procedural failures in agentic retrieval-augmented generation (RAG) systems, specifically focusing on instances where agents retrieve information but fail to inspect it before finalizing answers. By analyzing 12,000 paired trajectories from various datasets, the authors identify two distinct types of failures: pre-evidence discipline failures and post-gold-read failures, with a low overlap between them. Implementing a Read-Gate mechanism, which mandates reading after retrieval, significantly enhances accuracy, demonstrating that evidence-gathering should be treated as a separate control problem from reasoning.
Agents often skip critical evidence inspection, leading to significant accuracy drops鈥攆orcing them to read after retrieval can boost performance by nearly 20 points.
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.
