I hate to see that I never got a good score in validation because of too many irrelevant filps, I hope that the team may study how to prevent the bad flips, because even you are smart enough, you can never understand most of the flips in validation time.
As far as I understand, at the moment only the average result of the last 10 validations counts in the main % of successful flip solutions. That's why you shouldn't be afraid of losing the high % forever.
By the way, the team can't do anything with bad flips. This is decentralization, which works the way it should.