Unverified50% confidenceDecision RuleExact time
The gate uses only recall@5 as a hard threshold, with MRR and nDCG@5 as observational metrics, to avoid ranking fluctuations causing unstable builds.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedWhen recall@5 on adversarial questions is low, the correct action is to go back and check the chunking strategy; only after recall@5 meets the bar is it meaningful to invest in optimizing the embedding model or reranker.74% similarUnverifiedThe article uses Hit@5, MRR@5, retrieval latency, and failure modes as evaluation metrics.71% similarUnverifiedGate 5 requires that the benefits be substantial and the risks controllable, because the failure modes introduced by LLMs (hallucinations, uncertain retries, output format drift) cannot be completely eliminated and can only be contained with evaluation, monitoring, and human fallback.69% similarUnverifiedIncorporating retrieval quality metrics (such as recall@K) and latency metrics (P95, P99) into the same evaluation framework can avoid suboptimal decisions caused by single-dimension optimization.68% similarUnverifiedSetting the timeout below the observed p95 means deliberately giving up about 5% of calls in exchange for a bounded loop duration; whether this trade-off is acceptable depends on whether the task can tolerate skips.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1069283API
curl https://kongchang.com/api/v1/knowledge/claims/1069283MCP
get_claim(id=1069283)