7B local model prompt-based routing tested against Jev

Someone compared a token-saving routing approach that makes the LLM return only 1/0 or a confidence score with Jev on speed and results across several use cases.