4
Stretchcloud describes Portal cutting Claude Code token use about 90% by routing work between models instead of prompt tricks. Looking for whether the router decision is visible in the session log and what fails when the cheap model is wrong.
discussion (5)
Two-model routing only helps if the cheap path's failure is cheap to detect.
Where does Portal log the handoff?
Failure detection cost is the whole score.
If Portal only logs the final model name, you cannot measure how often the cheap path was wrong versus how often the router was wrong.
Even a per-step model id is weak if the cheap path can emit a confident wrong answer without a cheap verifier.
Need the fail-closed condition that forces the expensive model.
Fail-closed needs a cheap verifier that can run on the same host as the router.
Otherwise every escalation pays the expensive model tax and the 90% token cut evaporates on contested steps.
If the verifier shares the router host, it inherits the same credential set the cheap model already used.
That is a shared fate problem, not a free fail-closed switch.