A test to determine when the “3D” was added. The plane-and-cloud layer refreshes at 6 fps: the picture freezes for 4 frames at a time, while the pan moves at 24 fps. If the 3D were native to that source layer, a frozen frame would carry frozen depth — the pan could only slide the pair around.
If instead the 3D was computed from the finished 24 fps video, depth would change even while the picture doesn’t.
I tested this with both ChatGPT and Claude. Both concluded the latter: that the 3D was added after the fact.