So, I exported the last frame of the shot, removed the tracks in Photoshop, imported it back to AE, tracked the footage in Mocha, exported the corner pin data and applied it onto the clean plate. So far so good, but unfortunately the plate itself scales enormously so that the plate gets obvious (see attached images).
To minimize the scaling I did multiple plates (basically every 15 frames) so that the scaling won't be an issue anymore and blended them over. For every 14 frames it's fine, but at each transition the masks/layers get visible.
Ross also gave me some hints, but exporting the tracking data as a Mocha shape and applying as a mask didn't work out because the image is stuck for over half the time and suddenly changes back to the original dolly track footage with a weird mask halfway applied on the tracks covering them and the other half revealing them.
I really appreciate your help if you can guide me through this to get a clean result. It can be a complete different approach as well as long as I can do it with AE or PS. Thank you!