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

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.

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.

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