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

DevFeed: [Context as Infrastructure: Knowledge Compilation for Retrieval in Agentic Systems with Harnesses](<https://devfeed.tech/articles/context-as-infrastructure-knowledge-compilation-for-retrieval-in-agentic-systems-with-harnesses-112039.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.41172>)

Author: 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

Published: 2026-11-02T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [knowledge-engineering](<https://devfeed.tech/topics/knowledge-engineering.md>), [knowledge-management](<https://devfeed.tech/topics/knowledge-management.md>), [context](<https://devfeed.tech/topics/context.md>), [context window](<https://devfeed.tech/topics/context-window.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>), [AI-driven information retrieval](<https://devfeed.tech/topics/ai-driven-information-retrieval.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-systems](<https://devfeed.tech/tags/agentic-systems.md>), [artifacts](<https://devfeed.tech/tags/artifacts.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [bm25](<https://devfeed.tech/tags/bm25.md>), [compilation](<https://devfeed.tech/tags/compilation.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [database](<https://devfeed.tech/tags/database.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [gate](<https://devfeed.tech/tags/gate.md>), [governance](<https://devfeed.tech/tags/governance.md>), [governed-context](<https://devfeed.tech/tags/governed-context.md>), [graph](<https://devfeed.tech/tags/graph.md>), [harness](<https://devfeed.tech/tags/harness.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [integration](<https://devfeed.tech/tags/integration.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [latency](<https://devfeed.tech/tags/latency.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [systems](<https://devfeed.tech/tags/systems.md>)

## AI overview

The authors argue that agentic retrieval needs a knowledge compilation layer to control context, even when a capable search harness already achieves strong accuracy. In their runs, BM25/grep with MiniMax M3 answered 85% of BrowseComp-Plus and 91% of multi-hop FRAMES without compiled or vector indexes; structural artifacts showed no detectable FRAMES accuracy gain, and a PageIndex-style outline degraded at scale. A navigation tree with claim-evidence leaves reduced query tokens by 91% when evidence was sufficient and increased them by 1.9% otherwise, for a 38% average reduction at a measured sufficiency rate near 43%. A decision gate reduced latency from 140 to 30 seconds, while substitution can save tokens at the cost of more retrieval rounds and latency. The authors also describe other compilation artifacts as useful for deployment needs such as governance, integration, onboarding, temporal orientation, and continuity.

## Source excerpt

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.