Three rewrites of model selection

There is a question at the heart of Neo's architecture that I have answered four different ways in seven weeks. The question is: given a piece of code that needs to call an LLM, how does that code kno