A little more detailed explanation of the psastro measurement of the rotator center:

We start with a collection of images for a given altitude (and filter) taken at a series of rotator positions.

For each of these images, we use our static model of the chip positions relative to the reported boresite to determine the RA,DEC coordinates for the 0,0 pixels for each chip. The static model starts with the RA, DEC, and PA reported by the telescope for the boresite. It has a model for the location of all chips relative to that model. It has the ability to describe a model for the boresite motion as a function of rotator angle, but for now we keep the boresite fixed at the location -62, -139 pixels relative to the 0,0 pixel of ota33. The locations of all chips on the focal plane are described with linear offsets and potentially scales and rotations, but the current model keeps these at 1.0 and 0.0 respectively. The static model also ignores distortion.

The analysis then determines the astrometry for all chips relative to the external reference catalog (2MASS). We now know the actual RA,DEC coordinates for the 0,0 pixels (or any other) for each chip. We now would like to know for this single image, how well did the static model do at predicting the locations of the chips. To test the model, we don't really care about the predictions for the celestial coordinates; instead we care about the predictions for the locations of the chips in the local tangent plane (the projection relative to the boresite). We project the original RA,DEC values to the newly determined tangent plane. We then fit the measured tangent plane coordiantes to the predicted tangent plane coordinates with a linear transformation: this allows for an offset between the two systems, a rotation between the two system, and relative scale (and a shear term which we currently ignore). It also allows us to measure the scatter of the solution. I find that the scatter is fairly consistently in the range 5-8 pixels -- the large value is due to not applying a distortion model. The boresite offset (L,M) is a value defining the offset in pixels between the measured position of the boresite location on the sky as judged by the astrometry (and projected back onto the focal plate) and the nominal boresite position, assumed to be -62, -139 pixels relative to ota33.

We then take the rotation sequence of exposures and perform this analysis for each of them. We now have a sequence of L,M for different rotator positions. If the telescope's knowledge of the rotator position were exactly right, these values should all be the same. If the rotator were rotating about a different location (ie, a different location on the sky than that expected by the telescope), then these positions should follow a circular path with the radius given by the offset between the telescope-expected location of the rotator center on the focal plane and the true location of the rotator center. An offset of the center of this circle (or the zero-radius circle) would imply that the telescope has a different focal plane coordinate as a reference from the psastro analysis. We fit a general ellipse to this measured path. Below are figures showing the path for a series of altitudes. The bottom set of panels in each figure gives the residuals to the elliptical fit.