Variety engineering

Variety engineering is the design discipline concerned with deliberately matching — and sometimes mismatching — the variety of a regulator to the variety of the system it controls. Where the Law of Requisite Variety states the boundary condition (regulator variety must equal or exceed system variety), variety engineering is the practice of deciding how to satisfy that condition: by expanding the regulator's repertoire, by attenuating the system's perturbations, or by some combination of both. It is the applied counterpart to the theoretical law.

The discipline spans software architecture, organizational design, and policy engineering. In software, Information Hiding is a variety-attenuation engineering technique; polymorphism and generic programming are variety-expansion techniques. In organizations, Stafford Beer's Viable System Model is an architecture for engineering variety across recursive levels of management. In policy, regulatory sandboxes and sunset clauses are variety-engineering mechanisms: they modulate the variety of rules that firms must respond to, depending on the maturity and scale of the regulated activity.

The key insight of variety engineering is that the matching problem is not static. Systems evolve; their variety changes as they scale, adapt, and encounter new environments. A regulator that was well-matched at one scale may be under-matched at another. This is why Capability Control in AI safety is not a one-time design choice but a continuous variety-engineering process: as models scale, their capability variety grows, and the control mechanisms must be re-engineered to match.

Variety engineering is not about finding the right level of complexity. It is about building the feedback loops that detect when the level is wrong and adjust it. The systems that survive are not the ones with optimal variety. They are the ones with optimal variety-adjustment mechanisms.