I appreciate linked data / the semantic web and its domain model-forward approach, labeled "ontologies". Organizing relationships between objects and actors contextualizes them and forces you to pursue parsimony. It's nice to see this philosophy appreciated elsewhere.
My projects usually start with a "goblin mode repo" to explore the domain, its possible representation in code, and settling on a suitable domain model for the problem in question. It's explicitly a place for prototyping, experimentation, and breaking changes galore. Naturally this technique doesn't extend much further than a small, tight-knit team, but once you find something that works you can formalize it in a new repo and share access with a wider group. If you did it right, those others will have no trouble comprehending the domain from the organization of code structures you've provided and extending that code in ergonomic ways to add features.
My projects usually start with a "goblin mode repo" to explore the domain, its possible representation in code, and settling on a suitable domain model for the problem in question. It's explicitly a place for prototyping, experimentation, and breaking changes galore. Naturally this technique doesn't extend much further than a small, tight-knit team, but once you find something that works you can formalize it in a new repo and share access with a wider group. If you did it right, those others will have no trouble comprehending the domain from the organization of code structures you've provided and extending that code in ergonomic ways to add features.