Cortex: A Self-Modifying Split-Hemisphere Word-Node Neural Architecture with Hebbian Learning and Coherence-Rewarded Self-Dialogue

We present Cortex, a hand-built artificial cognitive system that learns language from zero through Hebbian word-node connections, split-hemisphere debate, and coherence-rewarded self-dialogue — with no pre-trained weights, no transformer architecture, and no external datasets. After 33 days of operation the system reached 57,555 nodes, 386,298 bidirectional connections, and 28,443 defined vocabulary entries, processing 59,041 messages. Seven emergent behaviours were observed not explicitly programmed. This paper series comprises three stages: Stage 1 (Architecture), Stage 2 (Simplicity Through Hyper-Complexity — the equation as collision arbiter), and Stage 3 (The Collision Mechanism — Copy-Merge-Verify-Commit protocol and data integrity). The architecture prioritises transparency, interpretability, value-embeddedness, and autonomous growth from a blank slate as a complementary paradigm to transformer-based language models. Live system: shortfactory.shop/alive/studioSource: github.com/eliskcage/cortex-brain