I've known for some time that the plate solving for PA is far more unreliable than from a preview image and suspected that it was being caused by sub-sampling of the image used for PA. Lately PA has been failing for me quite a lot after the 60 degree rotation which is a pain because it means I have to restart the PA from scratch.
I ran a few tests with a CCD simulator and it looks like the PA image sub-sampling is by a factor of 16 or thereabouts. Does it have to be so aggressive? Plate solving the full resolution image is lightning fast anyway.
