Unverified50% confidenceFactExact time
The HRR fix: store the original text vector instead of the bundle and mark it in the database as textv1 to identify and migrate old rows; for topic matching, the similarity between a paraphrased query and the stored text vector was +0.328, versus +0.009 when not matching.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen a retriever recalls documents based solely on semantic similarity, it may return both outdated and current documents simultaneously, causing the LLM to answer based on stale information.62% similarUnverified内容寻址存储通过内容哈希值寻址每条记录，任何篡改都会导致哈希值变化从而被检测59% similarUnverifiedMechanically splitting documents by a fixed token count causes headings and the paragraphs they belong to to be placed in different chunks, stripping the paragraph of key disambiguating information such as which package it belongs to59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1062254API
curl https://kongchang.com/api/v1/knowledge/claims/1062254MCP
get_claim(id=1062254)