EDCA-OS: A new expression-driven cognitive architecture for deterministic AI (github.com)

EDCA-OS (Expression-Driven Cognitive Architecture) is a newly published “AI operating framework” that claims to convert GPT-class models from probabilistic chat predictors into deterministic, auditable, expression-driven cognitive systems. Rather than relying on fine-tuning, hidden system prompts, plugins, or prompt-engineering tricks, EDCA-OS controls behavior through a stacked protocol architecture (L0–L5) and an expression language (Yuer DSL). Public-facing modules include a Structured Behavior Executor (SBE), Structured Retrieval Protocol (SRP) as a semantic-safe RAG alternative, a Structured Reasoning Override Engine (SROE) to correct faulty chains of thought, and an Externalized Memory Capsule (EMC) for traceable state. The repo offers demo evidence and conceptual docs but intentionally omits runtime code and sensitive internals. For the AI/ML community this is significant because it frames reproducible, verifiable multi-step reasoning and semantic defenses as system-level properties rather than prompt hacks. If the claims hold, EDCA-OS could improve deterministic chaining, reduce hallucinations via semantic-layer validation, enable long-horizon, recoverable tasks, and provide auditability useful for safety and compliance. Caveats: the published material is “public-safe” conceptual documentation—core implementations, scoring functions, and DSL grammar remain closed-source—so independent validation and adoption depend on forthcoming SDKs, visual specs, and demonstrable open interfaces.