Unverified50% confidenceBenchmarkExact time
In human verification conducted after freezing the pipeline, experts rated 0.995 of the conflicting tasks as "reasonable".
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe study is accompanied by a companion archive containing a frozen benchmark test set, evaluation audit logs, raw responses from the consensus experiments, and analysis code.66% similarUnverifiedIt is recommended to perform a lightweight integrity check before each checkpoint write to guard against silent loops.66% similarUnverifiedWhat determines the credibility of a public Agent score is whether the experimental protocol was frozen before the first scoring run, not model capability.64% similarUnverifiedIn the support-ticket routing comparison test, Laya answered correctly with probability 0.90, indicating that simple, clearly bounded classification tasks are easy to get right.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084334API
curl https://kongchang.com/api/v1/knowledge/claims/1084334MCP
get_claim(id=1084334)