It is important to understand how the planar tracker works to get most out of motor. The challenge here is to realise that this is a different way of doing tracking and that many of the restrictions you have with point trackers must be ‘unlearned’.
The tracker matches the area selected by the shape from one frame to the next, in a way similar to the way you would align a piece of tracing paper to the background image (i.e. by moving the tracing paper around until it matches)
The tracker calculates the motion between pairs of frames. It does NOT use a single reference frame. Hence you could let the tracker track a 100 frame shot, and although the tracker may inaccurately track from frame 25 to 26 it may still give you a good track from 26 to 100 - and all you have to do to fix the track for the whole should would be to go back to frame 25, try to adjust the shape and re-track from frame 25 to 26. You would not necessarily have to re-track the rest of the shot. This is probably not something you would expect. The reason is that, assuming only a relatively small error between 25 and 26, the shape would still be in a position where a majority of the pixels selected by the mask for tracking belong to the object being tracked.
The tracker does not know about the shape and it’s control points. It only sees the mask generated by the shape.
The best way to understand it is by enabling the ‘Grid’ overlay and playing back the tracked clip. The grid represents the plane tracked / the track itself. The shape sits on top of this tracked plane as is ‘pushed’ by the track. If you add a keyframe on the first frame and introduce a dramatic adjustment to the shape on the last frame, you will notice that the keyframed point travels smoothly across the grid between the keyframes. The track was not changed - the grid is still moving as before (changing the shape does NOT change the track directly) - the shape however is now keyframed and one or more of the points move around on the grid.