The Knox theorem.

Issue five is the technical issue. The argument behind why YG3 runs a content model and a reasoning model and a routing model instead of one big foundation model.

Specialization concentrates the loss landscape. A model trained only to write in one voice has fewer wrong answers available to it. A workflow that knows which model to call is more reliable than the largest model run alone.

The math is short. The implications are not.