Unverified50% confidenceDecision RuleExact time
In hybrid retrieval, about 50 candidates should be taken from each side rather than 5, otherwise if the correct document ranks 14th in one list, fusion will never see it.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBefore indexing ten thousand documents, manually sample fifty chunks first; if a person cannot answer the question from a single chunk alone, the splitting logic should be rewritten.70% similarUnverifiedWhen a team tests 1,000 combinations of strategy parameters, even if all combinations have zero predictive power, roughly 50 combinations will appear to perform excellently in backtesting purely due to randomness (the multiple comparisons problem).69% similarUnverifiedAnswer balancing requires that the yes/no labels of each question type be roughly matched in distribution, so that the expected accuracy of random guessing stays around 50%.68% similarUnverifiedCross-encoders are only suitable for re-ranking candidate sets (typically 20–100 items), not for full-corpus retrieval.67% similarUnverifiedThe article points out that the 50-sample evaluation set is too small, since gaining or losing a single sample can move accuracy by 2 percentage points, so the 86% result may simply reflect the model fitting these 50 samples rather than genuine generalization.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1072983API
curl https://kongchang.com/api/v1/knowledge/claims/1072983MCP
get_claim(id=1072983)