For Track 1, a token-efficient routing agent that decides local vs. remote per query to hit accuracy at the lowest token cost.
TranscendiantRouter answers tasks with a free local model and pays for an expert model only when the local model cannot agree with itself on an answer.