Sorry. Don't have the patience to read all the previous comments, but I agree with confirming scope height as well as possible mirage. Also, shoot at 100, 300, 500, 700, 900 yards or so. You may find that there is a scope height that works for all of those distances. I also agree with using a 100 yard zero. Just to minimize variables. Any chance of an updraft? Confirming your findings at different distances and also shooting different locations will help nail things down.
I've shot this load at various distances ~100 times, and the rifle/scope ~300 times, and another essentially identical rifle in 300WM ~1800 times. My skills generally behave. This load also has generally behaved, but a lot of those rounds were on steel in less than ideal conditions, or just groups looking for seating depth or velocity, so many accuracy issues weren't recorded as any more than an anecdotal piece of "data" in my own mind (unreliable).
This last effort was to really dial in for deer season, and I got a monkey wrench in the cogs.
I can deal with it, if by nothing else, keeping my Bambi murder to more "social distancing" friendly ranges.
As to the updraft idea, this occurred to me, particularly for the 2nd session last night and the 628 yard group's high offset compared to the 500 yd. It was halfway up a steep hill, and it's plausible some updraft was happening as the sun went down.
The mirage comments are also noted. Strings were slow enough and short enough (3 shots, cool, 3 shoots), that I doubt barrel mirage could account for what I saw on Monday, but there was also a little ground mirage at work, due to the sun coming in and out of the smoke.
I'm inclined to chalk this one up to multiple tiny, and nearly immeasurable tolerances, stacking in a bad way.
1/16 MOA up with mirage or wind patterns
1/16 MOA up with scope error I can't measure
1/16 MOA up with with my 200 yard zero
1/16 MOA from spin/jump affects
1/8 Rifleman error from the slight position change required to engage the 2nd target.
1/8 MOA up with random variation in the groups. Flip a coin (shoot) 100 times, and it's not hard to imagine you could find a string of 18 where high was over represented.
The above speculation accounts for a half MOA, that if conspiring in the "up" direction, could explain two groups on one day. That's why I like to shoot lots, and record my results (and not ignore misses or surprises as "outliers").