Good analogy. You’re gonna want the model to touch almost everything at some point.

I’d be interested in an effective security layer (e.g. not just allow/denying commands by some pattern) with a time-based component and perhaps some auditing features.