Stop Ranking Agent Configs by Average Score

Anthropic researchers argue that using average scores to rank agent configurations can be misleading. They suggest using alternative methods to evaluate agent performance.

1sec.ai