See https://photos.app.goo.gl/NKAPLvAuCPUKb9HT8 I tried to upload the .mov to phab, but it rejected the file format.

To reproduce this you need a page which is long enough that it doesn't fit in the window, and a TOC which is also long enough that it doesn't fit in the TOC pane without scrolling. In the exmample here, I'm using https://en.wikipedia.org/w/index.php?title=Wikipedia:Arbitration/Requests/Case&oldid=1116247536

In the movie, I'm continuously scrolling down through the TOC with the trackwheel on my mouse. As I do this, if I accidentally move the mouse so focus crosses the pane boundary (I'm using exaggerated movements here to make it more obvious), I switch from scrolling the TOC pane to scrolling the main pane. And once I do that, it jumps the TOC scroll position to match the main pane's scroll position, so the TOC jumps backwards. And if I then slide over a little bit so I'm back in the TOC, I pick up scrolling that pane, which is no longer in the place it was a moment ago.

It took me quite a while to figure out how to reproduce this, I suspect most people will think it's just scrolling randomly and have no idea why.

I think this is fundamentally the same problem as I reported in T318600. There's bi-directional coupling between the scroll position of the two panes, and small changes in one pane get amplified when the other pane tries to catch up. This is a classic example of small perturbations leading to large changes in observed state due to feedback instability.