This article introduces the concept of the institutional evolutionary moat — a structural regime of a collective in which the coherence mechanisms of a discipline (funding streams, publication channels, regulatory procedures, patient classifications) sustain its work with already constituted categories of objects, but by the same means block the constitution of new categories from within its own procedures. The regime is characterised not by active suppression but by procedural inexpressibility: the object of a program is not registered by the collective’s channels because procedural formats contain no place in which it could be submitted, evaluated, approved, funded, or presented as a claim — a layer that precedes the reputational work of marking described in STS and the Hess–Frickel coalition model of undone science. The article introduces a two-parameter matrix (rescuing variables × internal architecture), six operational criteria of categoriality with deductive justification of the number through six functional phases of the institutional processing cycle, a diagnostic protocol for independent application, and falsification conditions on two time horizons. The matrix is analysed on four empirical cases drawn from heterogeneous domains — R. Gatenby’s adaptive cancer therapy program (cell A), the Bukhara Protocol of eight AI systems (B), the bioelectric oncological program (C), and the LIGO Scientific Collaboration with its planned transfer to LIGO–India (D). The static typology is extended into a conditional-dynamic theory of regime migration between cells, yielding three falsifiable predictions. Limitations of the instrument are fixed in six layers, including the article’s self-reference with respect to its own formulation.
Anatoliy Kremenchutskiy (Thu,) studied this question.