Track, Rank, Crack: Epistemic Working Memory Scales Multi-Hop Reasoning in Language Agents
Ning Liu
Why It Matters
What makes this one worth your time
This work is relevant for AI researchers and engineers focused on improving the reasoning capabilities of language agents, particularly in complex multi-step tasks where maintaining a coherent investigative state is crucial.
SLEUTH enhances multi-hop reasoning in language agents by structuring their epistemic state.
Summary
The paper introduces SLEUTH, a structured epistemic working memory system for language agents that improves multi-hop reasoning by explicitly maintaining and organizing the agent's investigative state. This approach is tested across five multi-hop benchmarks, showing improved performance over existing methods, particularly as task difficulty increases. The paper also identifies and addresses the evidence sufficiency problem, demonstrating that structured reasoning organization is crucial for effective multi-hop reasoning.
Key contributions
- Introduction of SLEUTH, a structured epistemic working memory for language agents.
- Identification and partial resolution of the evidence sufficiency problem in multi-hop reasoning.
- Demonstration of improved performance on multi-hop reasoning benchmarks.
Notable insights
- Explicitly structuring an agent's epistemic state can significantly improve its reasoning capabilities.
- A lightweight commitment trigger is effective only when the agent's state is well-organized.
Possible limitations
- Not stated in the abstract
Abstract
arXiv:2607.12267v1 Announce Type: cross Abstract: Language agents that interleave reasoning and tool use degrade sharply as reasoning chains lengthen, even when each individual step is easy. We trace this to context dilution: an agent's investigative state (what it has confirmed, what it suspects, and what it still needs) lives only implicitly in a growing context window, where early discoveries are buried under later retrievals. We introduce SLEUTH, which makes this state explicit and actionable through a structured epistemic working memory: the agent maintains Confirmed Facts grounded to sources, Active Hypotheses ranked by evidence, and Open Questions that directly drive its next action. Across five multi-hop benchmarks and five established baselines, SLEUTH's advantage grows with difficulty, from +5 points on HotpotQA to +11 on 4-hop chains, surpassing Reflexion without multiple episodes. Analyzing where the remaining gap lies, we identify the evidence sufficiency problem: agents often find the answer but fail to commit, exhausting their budget on needless verification. A lightweight commitment trigger fixes this, but only when the agent already maintains structured state: the identical trigger applied to an unstructured agent yields no improvement, isolating organized epistemic state as the necessary condition for effective commitment. Finally, enforcing protocol adherence on a weaker model recovers up to +19 points on the hardest problems, showing that how an agent organizes its reasoning, not raw model capability, is the active ingredient for scaling multi-hop reasoning.