Recursive Requisite Variety: A Cross-Level Theory of Regulatory Capacity and Control Loss

Control is lost in structurally similar ways across systems that share no subject matter: an individual who can no longer check what a machine produced, a building whose automation degrades along the dimension it does not sense, a project organization that adds reviewers without adding scrutiny, and a workforce that delegates judgment to a system it cannot audit. These failures are ordinarily explained by four unconnected literatures. This article argues that they instantiate a common mechanism, and it offers that mechanism as a theoretical proposal rather than a tested account. Regulatory capacity is finite and must be allocated across scale, time, and dimension. Delegation does not remove regulatory load; it converts execution load into verification load, while the capability that discharges verification load is maintained largely as a by-product of execution and therefore decays under delegation. Because the allocation decision draws on the same bounded budget, the locally rational response to overload increases the load that produced it, and recovery requires a capacity injection exceeding the original deficit. The resulting structure is termed recursive requisite variety: the same allocation failure recurs at every level of a nested system, and a regulator’s effective variety is bounded by the effective rather than the nominal variety of the regulators beneath it. Seven falsifiable propositions are derived; none is tested here. The instantiations are conceptual recoveries with illustrative evidence rather than tests of the mechanism, and the article states at each level which of the two it is offering. Three of the propositions were not located, in the form stated here, in the scoping audit reported below: redundant regulators drawing on a shared source add effective variety only in proportion to their independent failure share; artefacts carrying neither recoverable provenance nor an independently checkable criterion impose verification load that no capacity can discharge; and overload is characteristically not registered by the actor experiencing it, because the delegate’s output exhibits the surface properties of reliable feedback.