I've always understood that when a GEM flips during a time series run one potential problem is that errors in a flat may show as an offset in photometry readings. The target remains near mid frame but the comp or check stars will flip sides in some orientations. A flat that isn't flat will cause altered readings on the comp and check causing an offset in the time series.
While I haven't normally seen this, recently I shifted to using another reduction package(I'll call M) and spotted an offset quite clearly. It took awhile to figure it out but it clearly happens at the flip. I checked the frames carefully and tried different combinations of check and comps, same problem.
As a test, ran the same images thru my other reduction package(I'll call C) and low and behold, no offset! I made several passes with each package matching the aperture sizes and comp stars, same result, one shows offset, the other doesn't. Some parts of the comparison I don't control of course. One package uses the Pinpoint plate solution to place the apertures, the other solves the plates during the reduction run to place apertures. The packages use diffent set up for apertures, M gives a finer control of the aperture sizes.
I then loaded 20 of the images to VPHOT, 10 on each side of the flip and reduced them there. VPHOt also does not seem to show an offset. The M package seems to show about 0.4 mag offset and that's just not there in C or VPHOT.
So the dilemma. Do I just pick the package with the results I like the best and blow off the other? I guess I'd prefer to understand why they are different, and if it's still some cockpit error I'm making, fix it. Is there anyother way to check these flats? I seem to remember moving a target star thru a grid pattern on the frame and making measurements at each point for comparison. Maybe that's next. Any other ideas?