Unverified50% confidenceBenchmarkExact time
Anthropic's experiments showed that combining embedding with BM25 and adding context to chunks reduced the top-20 retrieval failure rate by 49% (from 5.7% to 2.9%).
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAnthropic's experiments showed that adding a reranker on top of contextual retrieval brought the reduction in top-20 retrieval failure rate to 67% (from 5.7% to 1.9%).79% similarUnverifiedIn Anthropic's tests, the failure rate of Naive RAG (the correct text chunk not in the top 20) was 5.7%, dropping to 3.7% with contextual embeddings, 2.9% with contextual BM25, and 1.9% with reranking added.78% similarUnverified21.2% is a relative reduction rather than an absolute one, meaning the new version eliminates approximately one-fifth of the failures relative to the earlier version's failure rate baseline.73% similarUnverifiedModule residual anchoring lowered the hit rate under natural questioning from 28.7% to 22.0% while also reducing the measured behavioral drift, a trade-off of exchanging recall for preservation of non-target behavior.70% similarVerifiedIn long-chain tasks, if each step has an error rate of approximately 5%, the overall success rate may drop to below 36% after 20 steps.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1065182API
curl https://kongchang.com/api/v1/knowledge/claims/1065182MCP
get_claim(id=1065182)