Unverified60% confidenceFactExact time
In RAG systems, retrieval results are typically sorted by relevance score and concatenated in order, meaning the most relevant chunk is not necessarily placed in the most prominent position.
2
Sources
60%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
Unverified在RAG系统中检索结果按相关性分数排序拼接，最相关chunk不一定排在最显眼位置，可能因位置不当被模型忽视78% similarUnverifiedThe bottleneck of a RAG system often lies not in retrieval itself, but in how to tell the model via the prompt how to utilize the retrieved content.72% similarUnverifiedThe answer quality of RAG depends largely on the retrieval step rather than on the model's capabilities themselves.71% similarUnverifiedRAG检索依赖语义向量相似度匹配，无法检索只存在于PR评论或工程师记忆里的隐性决策知识70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/561502API
curl https://kongchang.com/api/v1/knowledge/claims/561502MCP
get_claim(id=561502)