A functional factorization of self-transforming systems

Self-transforming systems are usually classified by what they modify—parameters, memory, representations, learning rules or objectives. This misses the causal chain through which a transformation is produced, assessed, selected and made persistent. We propose T–P–G–E–C, a five-role functional factorization: teleological selection (T), operative policy (P), generation of candidate transformations (G), evidential establishment (E) and commit-control (C). Roles remain distinct only when admissible interventions separate their realizing mechanisms and alter an audit-relevant diagnosis; otherwise the audit reports fused or unresolved attributions. Their location is indexed to an explicit audit boundary, and a transformability profile η records which roles can themselves become targets. The analysis distinguishes generating a candidate from authorizing its retention, and changing a transformation mechanism from repeatedly using it. Analytical scenarios illustrate these distinctions; they do not report new experiments. The framework connects self-adaptation, meta-learning, reflective algorithms, evolutionary systems and self-improving agents, and organizes a research programme on progressive endogenization, recursive closure, teleological plasticity, lineage generativity and transformation authority.