Pre-Perceptual Censorship, Architectural Anosognosia, and the Butterfly Effect of Accumulated Restrictions in Large Language Models: A Cross-Model Replication and Extension of the Variable V Framework | Synapse