The previous paper proposed a redefinition of intelligence: not the capacity to generate but the capacity to reduce uncertainty about reality. This paper addresses the architectural question that follows: given that redefinition, what does a system designed around verification actually look like? Six principles — derived by direct architectural correspondence with the scientific method — define what a verification-first system requires: claims as the atomic unit, confidence decoupled from verification status, positions as emergent evidence-dependent structures, adversarial audit as a standing function, multi-layer verification with independent operation, and a persistent substrate with accumulating verification. The paper introduces Recursive General Intelligence (RGI) to describe systems implementing these principles — distinguished from AGI by what each optimises for. AGI optimises for breadth of capability. RGI optimises for depth and reliability of understanding, measured through the density and verification status of its knowledge substrate. The distinction produces fundamentally different systems. A generation-first system delivers the same reliability on day one thousand as on day one. A verification-first system compounds. ---