How does Geant4's object-oriented design specifically enable swapping physics models without rewriting detector geometry code?

I appreciate the specific technical question, but the provided preprints don't actually address Geant4's internal object-oriented architecture or the design patterns that decouple physics models from geometry code.

What *is* in the archive: Geant4 is described as "an object-oriented toolkit" [AG-2024.05-1380] whose design principles made it transformative for particle physics, and it's widely used as a reference standard for detector simulation. Several papers show that machine learning surrogates are now being trained to *replace* Geant4 calls for speed [AG-2025.09-1622], and some frameworks integrate custom physics packages (like dark matter modules) into Geant4 [AG-2024.01-1120]. But none explain the actual class hierarchies, virtual methods, or factory patterns that enable swappable physics models without rewriting geometry.

To answer your question properly, you'd want Geant4's design documentation or a paper that walks through its software architecture—perhaps a methods paper on the toolkit itself rather than applications built on top of it.