The Shape of the Problem: Domain-Invariant Structural Signatures in Activation Space Enable Cross-Domain Solution Transfer

INLP erases domain signal to near-chance while shape classification holds at ≥95.6% across 0.5B–7B on Qwen 2.5. Three transfer tests confirm structural signal is functional: cross-domain shape classification, nearest prototype, and strip-and-rehydrate. Subspace-targeted stochastic resonance within the INLP domain subspace finds domain signal standard INLP cannot reach. Provides a clean double dissociation confirming operational separation of domain and structure subspaces.