Metafunctional consistency is the selection principle in the evolution of modelling. Models are selected for (adopted, used, believed) and selected against (rejected or ignored) on the basis of ideational, interpersonal and textual consistency. There is a biological basis to the origin of this process to the extent that consistent modelling, in terms of construing experiences and assessing their relative value and relevance, increases the survival and reproductive prospects of the modellers.[1]
The evolution of models is a process of each fitting, adapting to, other models (via their perceivable expressions) with which they interact (are correlated).[2] Models are adaptations to other models.[3] As each model adapts to others, the environment to which other related models adapt changes, so that the evolution of models is a continual pursuit of the moving target of fitting.
So the sense in which the categorisable selects certain models over others is as follows. The categorisable (which includes categories of the categorisable) selects certain categorisings over others, and each model is built from categorisings that fit each other, in a process of each model adapting to similarly constructed models.
Footnote:
[1] Note that natural selection favours not just genes for phenotypic traits but also genes in organisms whose learnt behaviours result in more offspring, even though those behaviours are not the direct expression of genes. Cf Baldwin Effect.
[2] Just as, in biological evolution, gene complexes adapt (via phenotypes) to other gene complexes with which they interact.
[3] Even though a model is not the model, models are not arbitrary in the sense that any is as good as the next, since not all models equally fit others with which they are correlated (interact).