Six layers, one scroll, and a train. The hills behind you drift slower than the trees beside you, and that difference is the whole illusion of depth. Every ridgeline here was traced out of a 2.25-second recording; every speed below was measured from it. Scroll back up and the train reverses — it is a function of where you are, not of what time it is.
Mask alignment, reproduced at two independent scales.
0.85× and 0.87× from two runs. The viaduct rides this layer.
1501 px/s train against 419 px/s scroll in the recording. Here it is bound to your scroll, so it runs left → right as you go down.
A smooth gradient has no stable mask. Estimates ran −0.03 to 0.63, so we don’t claim one.