Interior of Inference
Capability study
The question
What is a model from the inside?
An informed external map of computation (weights, activations, logits—not ordinarily visible to a deployed model) and a spark of recursive self-modelling, left open. The reply lands on word—closing “word by word.”