When Should JEV Judge an LLM Response? A Practical Routing Rule

Evaluating LLM output at scale creates a familiar trade-off: a strong reasoning judge can catch difficult mistakes, but calling it for every response adds cost and latency. A recent study of JEV-as-a-