I do not know, but my best guess would be an improved post processing algorithm was introduced.

Wouldn't you just post-process all images again at that point?

I believe it's because the star's brightness does fluctuate, so activity cycle based?

Not stellar activity — a flare wouldn’t be visible on this scale of image. The main concerns are the observatory —- how aligned all the many dozens of optics are — and the earths atmosphere on the nights the target was observed.

In 2017 we captured five nights worth of images and averaged them. Some early years only had one night.

Around 2019 we also figured out some better observing strategies— all the later data is taken without a coronagraph. Turns out the coronagraph was hurting more than it helped, close in. Not to say coronagraphs aren’t useful, but some aspects of the observatory control software aren’t in place to get the benefits out of the coronagraph.