The problem
A single model for every request is easy to ship but expensive to operate. Rules catch obvious cases, while a general LLM is often too slow and too costly for a small routing decision.
The Jev question
Ask Jev which route fits the request: fast, balanced, or deep. Keep the options stable and define what each downstream model is allowed to do. Pin jev-1.13.0 once the policy is tuned.
The production loop
Log the Jev route, confidence, final answer quality, and any fallback to GPT or Claude. Revisit the question when your traffic or model catalog changes.