Hi @NDietz,
The situation is still somewhat complicated in that different software do still treat .tiff data differently.
The summary is that cryoSPARC reads .tiff data in the order that it is written on disk (i.e. cryoSPARC does not flip the y axis of incoming .tiff images) while other programs eg. MotionCor2 do flip the y axis of the raw data when loading it. This is the reason that in cryoSPARC, the gain reference needs to be flipped when it does not need to be flipped in other software, or vice-a-versa.
None of this affects the results though. Unfortunately I’m not completely confident about what all the particular tools/software that you mentioned (K2, imod, SerialEM) do internally, so I can’t give a definitive answer, but in cryoSPARC it is very easy to just import a few micrographs (say 10) with the gain reference, and keep re-running the import job with different flip/rotate settings until you see that the images outputted by the job (after gain correction) are correct.
Hope this helps!