Architectural choices in machine cognition are typically justified post hoc: a working system is built, performance is measured, and the design rationale is reconstructed from the result. We argue for the inverse — that the principal architectural decisions of a distributed cognitive system can be derived in advance from the computational organization of the mammalian neocortex. At the core of the DCA: atomic WMC-Agents (World Model ↔ Memory Controller coupling) composed fractally into multi-agent hierarchies — a pattern that recurs across cortical, hippocampal, and thalamic subsystems: a flow-producing processor coupled with a gating, filtering, and routing controller, the two locked in a recurrent loop whose attractors implement perception, action, and reasoning. The companion paper, Theory I, develops the formal convergence theory; this paper develops the biological substrate from which that theory derives its structure.