I see, so basically assume the two end points are at zero and there is some rotation accounting for the endpoint offset in real space. It still doesn't seem fu

I see, so basically assume the two end points are at zero and there is some rotation accounting for the endpoint offset in real space. It still doesn't seem fully accurate as I can imagine a non-rotated cubic curve with endpoints at an offset, but I assume your simplification works well enough.