Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
(Generation, ReferenceAnswer, Rubric) -> Score
Grade a target model's output using a separate referee model prompted with a grading rubric, target output, and optional reference answer.
Problem it solves
Programmatic metrics like exact match or BLEU fail to capture semantic accuracy in open-ended generations, while human evaluation is slow and expensive.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.