SoftModel: A Neural Model That Grows Its Own Topology — Governed Structural Growth for Continual In-Service Learning

Today, a neural system is almost always used in two phases — trained, then deployed — and in that regime it freezes twice: training ends, and the topology itself was never a degree of freedom. We take the opposite premise as an axiom — total plasticity: no part of a model, including its structure, is ever frozen — and derive the governance a lifelong learner then requires. The design's target regime is continual, in-service learning: a long-lived model on a non-stationary stream, whose stability is obtained from governance rather than immobility and whose capacity follows demand. The result is a growable soft model: an algebra of structural operators (width, hierarchy, composition, input interface, grown cycles, attention heads), each exact at the moment of application, each budgeted and audited, with adoption decided solely by a held-out reality gate that treats parametric and structural change uniformly. Local solving loops — grown directed cycles under an enforced contraction certificate, and self-processing units that refine newborn structure against a local objective — extend the same governance beneath the global pass. A complete from-scratch system realizes the whole account; its factory surface is operated end-to-end by a production LLM through a self-documenting tool surface. Two conclusions follow from the axiom by construction. Stability under lifelong change becomes an audit property of the lifecycle, not an immobility property of parameters — nothing need be frozen to be safe. Structure that follows demand removes the silent cap a fixed topology places on later capability where the capacity floor binds: a model born small need not remain bounded by its birth. A third is measured: in the worlds where this was measured, the marginal value of new capacity was unobservable before adoption, so workable growth governance took its ex-post form — speculation free, adoption earned on held-out reality, and silence in an unchanging world a governed outcome with a price. The same governance extends to evaluative signals — structure preference and policy optimization on the growing model under one gate — and the core method is evaluated on standard continual-learning benchmarks, where governed growth preserves the ability to keep learning along long task sequences and converts announced review into recovered competence. A pre-registered experimental program adjudicates the mechanism and value claims on the tested problems and reports its failures at full prominence; the map — positive and negative — is the contribution. Notes: 99 pages, 16 figures, 12 tables, 142 references. Pre-registered empirical program; negative results reported in full.