cs.IR提交于 2026-11-02已译

上下文即基础设施：带 Harness 的智能体系统中用于检索的知识编译

Context as Infrastructure: Knowledge Compilation for Retrieval in Agentic Systems with Harnesses

Zhichang Yu · Baining Cao · Sbaaoui Idriss · Qi Wang · Gangqiang Xu · Ling Qin · et al.

中文摘要

检索增强生成正走向智能体化：搜索 harness 会写查询、读结果、改写并决定何时停止。这基本解决了准确率问题——在我们的运行中，仅用 BM25/grep 和 MiniMax M3 的 harness 在 BrowseComp-Plus 上答对 85%，在多跳 FRAMES 上达 91%，无需编译索引或向量索引。但它暴露出上下文问题：缺乏约束的循环会重复查询、重读文档，并用未使用的段落填满窗口。缺失的一层是知识编译：可配置的摄取流水线，由 LLM 为文档补充单个分块无法包含的上下文——带 claim-evidence 叶节点的导航层级、实体图、时间线、思维导图、本体、wiki 与记忆。编译不是更好的检索器：在胜任的 harness 下，结构化产物在 FRAMES 上未带来可检测的准确率提升，PageIndex 式大纲在规模上反而退化。它是供给层。以 claim-evidence 对为叶节点的导航树——我们测量的机制——在证据充分时将查询 token 削减 91%，不充分时增加 1.9%，平均减少 38%；这意味着 2% 的盈亏平衡充分率，而实测接近 43%。决策门将延迟从 140 s 降至 30 s，但替代带来的 token 节省以检索轮次为代价，并自带延迟成本。其他产物——本体、图、wiki、时间线、思维导图、记忆——在部署中（治理、集成、上手、时间定向、连续性）产生价值，而准确率基准看不到这些。我们描述 RAGFlow 的流水线与 harness（开源地址 https://github.com/infiniflow/ragflow），它将循环转为五个受约束的站点，并让进入模型的上下文保持最小充分集；二者共同构成智能体原生数据库：向智能体交付已编译、受治理的上下文，而非排序后的段落列表。

关键要点

01仅用 BM25/grep 与 MiniMax M3 的 harness 在 BrowseComp-Plus 达 85%、多跳 FRAMES 达 91%，无需向量索引。
02无约束检索循环会重复查询、重读文档并用未使用段落填满上下文窗口。
03claim-evidence 导航树在证据充分时削减 91% 查询 token，不充分时增 1.9%，平均降 38%。
04决策门将延迟从 140 s 降至 30 s，但 token 节省以更多检索轮次和延迟为代价。
05本体、图、wiki 等产物在治理、集成、上手等部署场景产生价值，准确率基准无法体现。

解读

尚无解读。

原始英文摘要

arXiv:2610.41172v1 Announce Type: cross Abstract: Retrieval-augmented generation is becoming agentic. A search harness writes queries, reads results, reformulates, and decides when to stop. This largely solves accuracy: a harness with only BM25/grep and MiniMax M3 answers 85% of BrowseComp-Plus and 91% on multi-hop FRAMES in our runs, without compiled or vector indexes. It exposes a context problem: an undisciplined loop repeats queries, rereads documents, and fills the window with unused passages. The missing layer is knowledge compilation: a configurable ingestion pipeline where LLMs enrich documents with context no chunk contains alone--navigation hierarchies with claim-evidence leaves, entity graphs, timelines, mind maps, ontologies, wikis, and memory. Compilation is not a better retriever: under a competent harness, structural artifacts gave no detectable accuracy gain on FRAMES, and a PageIndex-style outline degraded at scale. It is a supply layer. A navigation tree whose leaves are claim-evidence pairs--the mechanism we measure--cuts query tokens by 91% when evidence suffices and adds 1.9% when it does not, a 38% average reduction; this implies a 2% break-even sufficiency rate and a measured rate near 43%. A decision gate cuts latency from 140 s to 30 s, though substitution's token saving is bought with retrieval rounds and carries its own latency cost. Other artifacts--ontology, graph, wiki, timeline, mind map, memory--pay off in deployment (governance, integration, onboarding, temporal orientation, continuity) where accuracy benchmarks cannot see them. We describe RAGFlow's pipeline and harness (open source at https://github.com/infiniflow/ragflow), which turns the loop into five disciplined stations and keeps the context entering the model the smallest sufficient set; together they form an agent-native database: a store that hands the agent compiled, governed context instead of a ranked list of passages.