On the pre-filter audit sample, all three validation judges exceed 90% agreement and Cohen’s kappa of 0.80 with adjudicated human decisions. Their acceptance rates are below the human reference.

On the pre-filter audit sample, all three validation judges exceed 90% agreement and Cohen’s kappa of 0.80 with adjudicated human decisions. Their acceptance rates are below the human reference. · CiteArk

测量结果

已评估 0/21 项

获得支持 0有争议或冲突 0尚未评估 21

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
comparison: two primary annotators before adjudication	96%	—	尚未评估
comparison: two primary annotators before adjudication	0.916 dimensionless	—	尚未评估
evaluator: Human reference	62%	—	尚未评估
evaluator: MiniMax-2.5 · reference: adjudicated human labels	58.5%	—	尚未评估
evaluator: MiniMax-2.5 · reference: adjudicated human labels	91.5%	—	尚未评估
evaluator: MiniMax-2.5 · reference: adjudicated human labels	0.823 dimensionless	—	尚未评估
evaluator: MiniMax-2.5 · reference: adjudicated human labels	95.7%	—	尚未评估
evaluator: MiniMax-2.5 · reference: adjudicated human labels	90.3%	—	尚未评估
evaluator: MiniMax-2.5 · reference: adjudicated human labels	92.9%	—	尚未评估
evaluator: GPT-5.5 · reference: adjudicated human labels	60%	—	尚未评估
evaluator: GPT-5.5 · reference: adjudicated human labels	92%	—	尚未评估
evaluator: GPT-5.5 · reference: adjudicated human labels	0.832 dimensionless	—	尚未评估

On the pre-filter audit sample, all three validation judges exceed 90% agreement and Cohen’s kappa of 0.80 with adjudicated human decisions. Their acceptance rates are below the human reference.

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果