Unified Architecture v1.2 (EPS-Integrated)

Unified Architecture v1.2 (EPS-Integrated) defines a complete, modular technical architecture for human-centered AI systems designed around cognitive safety, transparency, and regulatory compliance. The document specifies the internal structure, data flows, and enforcement mechanisms of the LUMEN system, integrating three cross-cutting EPS compliance layers: Environmental (E): computational cost and energy tracking Privacy (P): privacy-by-design ingestion, anonymization, and redaction Security (S): tamper-evident, hash-chained audit logging At its core, the architecture enforces human agency as a non-negotiable principle, separating analysis, option generation, and decision-making into explicit stages where the system may structure information but never direct outcomes. The reasoning engine is constrained to descriptive analysis and bounded option generation, with all choices remaining under explicit human control. The specification details: Layered system design (interface, ingestion, processing, reasoning, feedback, deployment) Reality anchoring and fact/assumption separation Guardrail enforcement at every processing stage EPS-compliant logging, reporting, and export formats Cryptographically verifiable audit trails Deployment scenarios spanning research, enterprise, consumer, and education contexts This document is written as a production-ready technical specification, suitable for academic publication, implementation guidance, and regulatory review. It includes explicit alignment with EU AI Act requirements, GDPR principles, and ISO/IEC AI and security standards, and is intended to function as auditable evidence of system-level compliance rather than a policy statement or conceptual proposal.