Dodgy aperture photometry calibration on data before MJD=58000

[JamesMunday98]

Hi. I've been extracting aperture (not difference imaging) photometry with atlas for a few hundreds-thousands of targets and have noticed that quite often the data before MJD~=58000 is very different to the data proceeding it. A good example is this star. It can be found at 273.26941194790476 +22.41385305621907 and is fairly well isolated. The data appears to droop towards lower magnitudes (as is common in multiple, but NOT all, targets isolated by at least 20 arcseconds with no very bright star within 1 arcminute) but then appear to become well calibrated after. See the plot below, colours are unique filters

Do you have any idea what the cause of this could be? All stars I am entering include proper motions in mas/yr from gaia dr3 with its reference epoch. I can send more examples if that would be useful - I have not been able to spot a trend yet of certain declinations where this is more impacted.

Thanks