Unverified50% confidenceSolutionExact time
Hierarchical retrieval can be extended into a three-level structure of corpus-level → document-level → passage-level, controlling computational overhead while maintaining recall.
1
Sources
50%
Confidence
Long-term
Relevance
9/6/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCLAUDE.md is loaded across three levels—user-level, project-level, and directory-level—using an overlay merge mechanism in which more specific scopes have higher priority.67% similarUnverifiedThe project adopts a three-layer architecture design consisting of a data collection layer, an analysis and processing layer, and a presentation and push layer.67% similarUnverified父子块策略将文档同时索引为大块和小块，检索时用小块精准定位，生成时返回大块提供完整上下文65% similarUnverified在RAG链路中PDF解析是最上游环节，解析输出的文本质量直接决定向量化、检索和生成三个下游环节的天花板63% similarUnverified编译式RAG通常采用三层架构：原始层保存文档原始内容用于溯源，编译层进行结构化抽取和关联建模，应用层面向问答检索支持带溯源的回答62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/864516API
curl https://kongchang.com/api/v1/knowledge/claims/864516MCP
get_claim(id=864516)