The Concept Collider: Fracture-Targeted Creative Emergence as a Glass-Box Computational Framework

Title:The Concept Collider: Fracture-Targeted Creative Emergence as a Glass-Box Computational FrameworkAbstract:Computational creativity has split into two eras: an interpretable formal era of limited reach, and a generative era of vast reach but opaque, black-box behaviour. The Concept Collider sets out to recover the formal era’s interpretability within the generative era’s reach. It models creative emergence as a precision-targeted collision between two concepts from distant domains: it locates a concept’s deepest structural tension—its fracture—and projects onto it the resolution another concept supplies elsewhere. We make three contributions. First, a creativity framework grounding projectile selection in distributional semantics (Cf ∝ dW × S × vrel × I × Φ(E)), a proportional model whose two core geometric factors—semantic distance dW and fracture homology S—are instrumented (dW computed, S LLM-estimated), the other three modelled but not yet measured. Second, an operational four-step protocol (STRETCH → MAP → TARGET → COLLIDE), with two implementations, validated on a clean corpus of 424 collisions (52 of them independently scored) together with a designed battery of adversarial and positive controls that separate productive from degenerate conditions. Third, because every emergence is fully traceable—source, fracture, projectile, mapping—the Collider is glass-box at the level of process. This makes it an architecture designed to serve as an auditable instrument, and, we argue, a first building block toward a science of creativity: a collision’s validity is read from its provenance chain rather than from its output score, a point its degenerate inputs make sharp, since fluent output cannot certify validity. This is the extended version of [Wahl, 2026].