Unified Cognition from a Single Optimization Principle: Active Inference in MeTTa with Emergent Reasoning, Planning, and Self-Knowledge

A cognitive architecture targeting artificial general intelligence in which perception, action selection, causal reasoning, temporal planning, self-knowledge, and conversation all derive from a single optimization principle: minimize Expected Free Energy (EFE). Implemented in approximately 12,000 lines of MeTTa, validated through executable Python specifications, a pure Python MeTTa evaluator, and a stateless LLM orchestrator running end-to-end with a local language model. Three forms of inference — induction, deduction, and abduction — operate as hypothesis generators under uniform metabolic selection. Code and full documentation available at https://github.com/Jolnir-Sefir/dagaz