A dark-lucid reinforcement learning architecture that learns, dreams, and acts through latent world models, causal verification, and memory-anchored adaptation under uncertainty.

A method-neutral protocol for recording and verifying evidence that latent model states causally influenced downstream decisions, outputs, or actions, without requiring storage of raw internal activations.

End-to-end prime factorization in a generative LM. 40M-param GPT that learns algebraically verifiable prime-factor signatures at negligible language cost (+1.7% PPL). Paper (Zenodo) + triadic-head (PyPI) + reptimeline.

Does your AI agent actually follow rules? 13 pre-registered experiments + 5-layer verification architecture. Paper, data, code — all public.