Dreams as Predictive Neural Rendering: Memory Recombination, Internal Simulation, and State-Dependent Testing

This theory-and-methods paper develops a source-faithful Self Aware Networks account of dreams as state-conditioned predictive neural renderings. It separates scene generation, source attribution, vividness, memory recombination, conscious access, and later report; specifies discriminating experiments and failure conditions; and supplies a reproducible synthetic reference application with retained adverse and null results. The application result is bounded to its declared generator-aligned development setting and is not evidence that human dreams have been decoded or that the proposed biological mechanism has been validated.