DeepEval added a mode where Jev, not a generative LLM, decides the score. What that buys you, what the real benchmark numbers show and where it runs out of road.
Everyone asks this question. Most of the answers online are wrong or at least incomplete.