Topology or Metaphor? - Formal Obligations for Models of Language and Subject Development

Topological vocabulary appears in models of neural activity, semantic organization, linguistic structure, and development. Its inferential content depends on the object to which topology is assigned. This position paper proposes a warrant architecture linking observations, a representa tion map, a topology-bearing construction, transformations, invariants, estimators, construct bridges, and comparison models. Several separation results expose the cost of omitting that architecture. Input–output behavior remains unchanged after an observationally silent topo logical factor is added. Recognition of unbounded bracket nesting and nontrivial fundamental group vary independently. A fixed contractible carrier supports arbitrarily many category labels. Homeomorphic state spaces can support dynamically inequivalent maps, while a bifurcation can occur on a fixed ambient topology. A positive construction is also supplied. Intervention-indexed response laws induce a pseudometric on developmental histories; its zero-distance quotient is a metric space whose finite samples admit a Vietoris–Rips filtration. Uniform pairwise metric error of at most δ yields a δ-interleaving and bottleneck error at most δ. The result licenses a stable topological summary of a declared response representation. It does not identify an innate grammar, a neural carrier, knowledge, or a subject. The paper concludes with model-specific obligations for recursion, categorization, cross-language comparison, developmental transitions, and longitudinal validation. The position is constructively conservative: topology can become explanatory, while the representation and attribution bridges remain part of the theory.