Unverified50% confidenceDecision RuleExact time
When evaluating the speedup of MoE speculative decoding, both the number of accepted tokens and the number of unique experts should be reported, to avoid the optimistic bias of a single metric.
1
Sources
50%
Confidence
Long-term
Relevance
10/29/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedCalibration concerns whether the probability values are trustworthy, while discriminative power concerns whether high-confidence samples are more likely to be correct than low-confidence samples; even if calibration is biased, as long as confidence maintains ranking discriminative power, it can still be used as an auxiliary variable for stratified sampling.70% similarUnverified'Cherry-picked results' means researchers report only the results from among multiple experiments that support their hypothesis, which is a form of selective reporting.70% similarUnverifiedIf monitoring covers only final business metrics, the small traffic sample in the canary phase may be insufficient to trigger a statistically significant alert, so more sensitive proxy metrics such as prediction distribution and feature distribution should also be compared, with reasonable sample sizes and confidence intervals set.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1110416API
curl https://kongchang.com/api/v1/knowledge/claims/1110416MCP
get_claim(id=1110416)