教程 / 实战普通
The Eval Pyramid, Part II: Measuring Agent Reliability with pass@k, pass^k, and Judges You Can Trust
内容摘要
One passing trial says a system can succeed. Repeated trials, computed per task, say whether people can depend on it. Here's how to read pass@1, pass@k and pass^k, why the same 75% can describe two opposite systems, and how to check a model judge before it gates a release.