Simple setup. Sop level. Two stream merged in one named "test":
1)Random geometry, uv, normals, packed, added unique "path" attribute (s@path = '/root/somename/')
2)Random geometry, uv, normals, packed, added unique "path" attribute (s@path = '/root/somename1/'), transform node and animated some parameters.
Lop level.
"Sop Import" imports "test".
"Load Layer For Editing" loads alembic file created from "test".
"Sop Import" - time dependent - Yes, and "push" this behavior down the stream. Sampling geometry every frame and forcing recalculation in other nodes. Playing animation as intended.
"Load Layer For Editing" - time dependent - No. Node calculates one time and just pass time sampled attribute down the stream. (I don't go deep in this, for me, it's like attribute has value not for frame, but array for all frames. But now its doesn't matter). Playing animation as intended.
This situation means that I can Import animated transform to Lop without time dependency in node.
When I export to USD "Load Layer For Editing" output, it's pretty good: two geometry, points positions, st, normals, one time sampled transformation.
When I export to USD "Sop Import" it's pretty bad: two geometry, points positions, st, normals, and for both time sampled transformation, visibility, extent. Literally for 1000 frames program generates:
token visibility.timeSamples = { 1: "inherited", 2: "inherited", ...... 1000: "inherited",}
Questions:
1)How I can get rid of time dependency in "Sop Import", and don't use alembic? Use Cache Lop, or exists more elegant solution?
2)How you add materials to animated streams? Or more correctly, how you more efficient adding materials to animated streams?
3)How I can control which parameter get time samples, and turn of this for visibility, extent or others attributes?
Why I ask this? Q1 and Q2 simply for speed up Lop workflow, because time dependency in lop nodes its big pain. Use reference and create new USD for materials looks strange for me.
Q3 I need export animation to USDZ, and get twice larger file size, because Houdini export visibility and extent, not a good idea.
I think I miss some information, and exist more efficient way for export transformed geometry and not save this to alembic
Of course I can create two "Sop Import", for static and animated geometry. But this only help to not generate unnecessary time samples for part of entire scene, and alembic solution can separate geometry without me.


