Glossary: Evaluating LLM Apps
How much a score varies from one sample of cases to another. It is why 82% versus 87% on 40 cases proves little.
Confidence intervalStandard error