Most attempts to personalize an AI quietly collapse a person into a type. CMP makes the revisable guess the unit instead.
Most attempts to personalize an AI quietly collapse a person into a type. You take a quiz, you come out an INTJ or a 'Driver' or a colour, and from then on the system treats the label as the truth. That's the opposite of how working relationships actually form. You don't learn how to work with someone by assigning them a class — you make a guess, watch what happens, and revise.
The Cognitive Model Protocol is an attempt to make that revisable guess the unit, instead of the label. A CMP model is a set of hypotheses about how a person and an AI can collaborate. Each one carries a confidence (how sure we are right now), an uncertainty (what we still don't know), and a falsifier (the thing that would change our mind).
That last one is the whole point. A claim you can't falsify isn't a hypothesis, it's an identity statement — and identity statements are exactly what we're trying to avoid. 'You are analytical' can't be wrong. 'When things get messy, you'd rather I synthesize than ask one question at a time' can be wrong, and you can tell me so.