tylerart
Ok, I think I figured out that I have to reference the asset file directly into my stage, I can't just pipe into a new reference node with a new layer. Once I've referenced the file directly, it seems to work fine. I'm guessing there's a flag somewhere that I need to set, just not sure what I'm missing.
I'm not sure if this is the original thread/question, but your issue is related to composition strength ordering LIVERPS: the variant set opinions (i.e. "V") are weaker than the direct layer opinions (i.e. "L"). When you defined the sphere, you authored an active opinion. When you referenced the asset into a scene, that weakened the activation opinion on sphere1 down to an "R", and therefore the "V" opinion won out.
Also worth mentioning, USD does not allow animating Activation or Variant Sets selections. Sometimes Solaris gives the illusion that it can, because you can animate a LOP node, and it recooks and sets activation as the time changes. But when you write out your animation to disk, you'll only see the last value authored. Whereas with visibility, you would see multiple time-samples animating the objects on/off.
tylerart
An additional thought - If you have a scope with prims inside them, and deactivate the scope, you can't re-activate the scope again to bring those prims back, despite the data being in the usd file. It seems that the data below the scope is no longer accessible?
Setting active on the prim should compose the children prims into the stage. I think the only time the children of inactive prims are fully and completely removed, is when you flatten the stage. Otherwise, the data is still there, but just not loaded into memory.
Hope this helps!