Monadic IO, Uniqueness types (as in: Clean), voluntarily working in a pure subset until the End of the World (Scala), effect-first languages, etc. We can impose

Monadic IO, Uniqueness types (as in: Clean), voluntarily working in a pure subset until the End of the World (Scala), effect-first languages, etc. We can impose actual semantics upon an ultimately imperative computer. Is it restrictive? Yeah, but that's, like the point, man. Restricting what can happen allows us to reason about what might happen given a piece of 'random' code. In an imperative-only world, Capabilities serve a similar function. The fact that the world is made up of elementary particles subject to the 'laws' of thermodynamics has very little impact on the semantics we choose for our programming languages. (If you want to take your appeal to 'nature' that far.)