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

Computer Science > Information Retrieval arXiv:2610.41172 (cs) [Submitted on 30 Oct 2026] Title: Context as Infrastructure: Knowledge Compilation for Retrieval in Agentic Systems with Harnesses Authors: Zhichang Yu, Baining Cao, Sbaaoui Idriss, Qi Wang, Gangqiang Xu, Ling Qin, Xiaoling Jin, An Liu, Shaohao Cai, Xiaoyang Bao, Shengle Chen, Changchang Ding, Pengfei Zhu, Jingxiang Xu, Yifeng Mao, Lu Liu, Hai Jin, Yingfeng Zhang.

计算机科学 > 信息检索 arXiv:2610.41172 (cs) [提交于 2026 年 10 月 30 日] 标题：作为基础设施的上下文：代理系统中检索的知识编译与工具链。作者：Zhichang Yu, Baining Cao, Sbaaoui Idriss, Qi Wang, Gangqiang Xu, Ling Qin, Xiaoling Jin, An Liu, Shaohao Cai, Xiaoyang Bao, Shengle Chen, Changchang Ding, Pengfei Zhu, Jingxiang Xu, Yifeng Mao, Lu Liu, Hai Jin, Yingfeng Zhang。

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.

摘要：检索增强生成（RAG）正变得日益智能化（agentic）。搜索工具链（harness）负责编写查询、读取结果、重构问题并决定何时停止。这在很大程度上解决了准确性问题：在我们的运行测试中，仅使用 BM25/grep 和 MiniMax M3 的工具链在 BrowseComp-Plus 上达到了 85% 的准确率，在多跳 FRAMES 上达到了 91%，且无需编译索引或向量索引。

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.

这暴露了一个上下文问题：无序的循环会重复查询、反复读取文档，并用未使用的段落填满上下文窗口。缺失的层级是知识编译：这是一种可配置的摄取管道，LLM 在其中为文档添加单个分块所不具备的上下文——包括带有“主张-证据”叶节点的导航层级、实体图、时间线、思维导图、本体、维基和记忆。

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%.

编译并非更好的检索器：在高效的工具链下，结构化制品在 FRAMES 上并未带来可检测到的准确率提升，而 PageIndex 风格的大纲在规模化时反而表现下降。它是一个供应层。一种以“主张-证据”对为叶节点的导航树（我们所测量的机制）在证据充足时可减少 91% 的查询 Token，在证据不足时增加 1.9%，平均减少 38%；这意味着盈亏平衡的充足率为 2%，而实测率接近 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.

决策门将延迟从 140 秒缩短至 30 秒，尽管替代方案节省的 Token 是以增加检索轮次为代价的，并带有其自身的延迟成本。其他制品——本体、图谱、维基、时间线、思维导图、记忆——在部署阶段（治理、集成、入职、时间导向、连续性）体现出价值，而这些是准确性基准测试无法衡量的。

We describe RAGFlow’s pipeline and harness (open source at this https URL), 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.

我们描述了 RAGFlow 的管道和工具链（开源地址见此链接），它将循环转化为五个规范化的站点，并确保进入模型的上下文是最小的充足集；它们共同构成了一个代理原生数据库：一个向代理提供已编译、受治理的上下文，而非仅仅是排名段落列表的存储系统。