OpenRouter in production: routing without romanticizing the model

Stub draft. Angle: cost, failover, and when a smaller/faster model is the right call for intake — escalate size only when the turn needs it. OpenRouter (or any router) earns its keep when routing policy is boring and measurable.

Coming: concrete failover rules, what I log per turn, and when “best model” is just an expensive default.

Contact if you want the full writeup sooner.