LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management

• 类型：arxiv
• 标识：2610.13075
• 链接：http://arxiv.org/abs/2610.13075v1
• 主分类：agent
• 形态：method
• TLDR：Corpus-level long-context reasoning requires agents to acquire and integrate evidence across document collections that exceed their context windows. Existing agent harnesses enable incremental access to these corpora, but acquired evidence and evolving conclusions often remain entangled in a growing interaction history. This implicit state management can lead to two coupled failure modes: unbounded exploration, where retrieval expands the interaction history without resolving decision-relevant information gaps, and evidence-state drift, where previously acquired evidence or subsequent correcti
• 副分类：evaluation
• 待LLM分类：否
• 标题中文：LongHarness：通过有状态证据管理推进 100M Token 语料上的智能体推理
• TLDR中文：语料级长上下文推理要求智能体在超出自身上下文窗口的文档集合中获取并整合证据。现有智能体 harness 支持对这些语料的增量访问，但已获取的证据与不断演化的结论常常纠缠在不断增长的交互历史中。这种隐式状态管理会导致两种耦合的失效模式：无界探索——检索不断扩展交互历史却未能解决决策相关的信息缺口；以及证据状态漂移——已获取证据或后续更正……
• 来源文件：
• /inbox/tom/_candidates/2026-10-12-agent-rag-longcontext-candidates.json