llm-as-a-verifier

A general-purpose verification framework that gives fine-grained reward and progress scores for any agent without extra training, supporting best-of-N selection, test-time scaling and progress tracking.