Governance-Enforced Cognitive Execution: The TACA Architecture and the Minerva Runtime

Modern AI systems are increasingly capable of generating proposed actions, plans, and operational decisions. However, most contemporary AI architectures implicitly grant execution authority to the same systems that generate those actions. This coupling between reasoning and execution creates a structural governance gap: systems can propose actions faster than organizations can authorize them. This paper introduces the concept of execution admissibility, defined as the determination of whether a proposed system action is permitted to execute under defined governance policy. We present the Tiered Autonomous Cognitive Architecture (TACA), a governance-first system design that separates cognition, governance, and execution into explicitly enforced architectural layers. Within this architecture, a cognitive engine may analyze tasks, generate plans, and propose system mutations, but it cannot directly execute them. All state-changing operations must pass through a governance enforcement layer that validates declared capabilities, records decisions, and authorizes execution through deterministic lifecycle controls. We further present Minerva, a working cognitive runtime operating within the TACA architecture. Minerva performs deterministic reasoning, capability analysis, and governed proposal generation while remaining structurally incapable of bypassing governance enforcement. System capability expansion occurs through a governed module lifecycle that requires explicit approval before new functionality becomes operational. The architecture demonstrates that governance can function as a runtime infrastructure layer rather than an external policy mechanism.