Centroids of the target change unpredictably near the end of a staring mode AOR. When examining the images by eye, the star is not actually moving around the detector. However, header values CRVAL 1 and 2 change at the boundary from normal centroids to random centroids.
This photometry code takes ra and dec as input and converts those to X and Y position using the header astrometry, and then uses those initial values of X and Y to refine an actual centroid. Pointing history files are what the SSC uses to generate the astrometry in the header given the telemetry from the telescope. In this particular AOR, the pointing history file which was used to reconstruct the astrometry changed at an SCLK time of ~1.06246E9. This caused a change in CRVAL in the headers. When CRVALs change, then it appears as though the image shifted, when in fact it didn’t. In this case it shifted by enough distance that the centroiding routine lost the star and is attempting to centroid on noise, which explains the large scatter in positions after the shift.
We have since re-run the pipeline on this AOR, and it now has correct astrometry in the headers. If you see something similar to this in your data, contact the SSC to correct the AOR. We have seen this problem only once in the warm mission.