The principle here is that an attribute x of an entity A is not explained by reference to a constituent entity B that has the same property. The strength of an arch is a property of arches, for example, not of the things from which arches are constituted.

That doesn't imply that theremust be a B in the first place, merely that whether there is or not, referring to B.x in order to explain A.x leaves x unexplained. (Of course, if there is no B, referring to B.x has other problems as well.)

I suspect the "top"/"bottom"/"level" analogy is misleading here. I would be surprised if there were a coherent "bottom level," actually. But if there is, I suppose the sign that I've reached it is that all the observable attributes it has are fully explainable without reference to other "levels," and all the observable attributes of other "levels" are fully (if impractically) explainable in terms of it.

At any level of description, there are observable attributes of entities that are best explained by reference to other levels of description, but I'm not sure there's always a clear rank-ordering of those levels.

"If atoms are really to explain the origin of color and smell of visible material bodies, then they cannot possess properties like color and smell."

I want to point out that in this post, you were quoting sediment quoting Hofstadter who was referencing Hanson's quoting of Heisenberg. Pretty sure even Inception didn't go that deep.