Foreman

Evals

Score the agent against a fixed suite: a programmatic citation-validity check plus an independent LLM judge.

Recent eval runs