> If it is non-physical then it cannot have any impact on the physical world by definition.

Well, that depends on your definition.

If non-physical things have no interface with physical things, then they might as well not exist -- their non-existence is tautological and the hypothesis is meaningless. So the only meaningful "non-physical" hypothesis is one that allows an interface with the physical.

> the only meaningful "non-physical" hypothesis is one that allows an interface with the physical.

I don't think that's meaningful, it sounds to me like a contradiction of terms. If it is non-physical but it can interact with physical objects as if it were physical, what is the label "non-physical" actually describing? In a world where there are physical entities but also non-physical entities with physical interactions, how is the non-physical entity distinct from the physical one in terms of observable reality?

> what is the label "non-physical" actually describing

In this case, it would be describing the consciousness phenomenon -- which we can't get at using normal, scientific, physical observations of the world.

If X exists, we all know it exists and can talk about it, but we have no way to observe it (in fact, all our observations are restricted to being through it) - then I'd venture we're on an edge of reality itself. In my opinion, a non-physical hypothesis here is allowable if it has more explanatory power than the alternative.