I wanted to post my statistics page, but did not find a "spoiler" feature to hide this unnecessary blunter from the innocent people.
For the sake of the argument...
OK I have looked at numbers. The main problem is your speed. To be 97% sure you need 25% deviation if I have calculate that correctly... Unless you have a log of shares.
Only block 18267 shows that and you gain on it 0,000323941. So others might have lost on that one just a bit or you were extremely lucky to get shares right at the end. Did you see any strange estimated rewords during that block?
Close to 95% is also a block 18264. You lost 0,000310308. But it was invalid and still a chance it could happen... So loss is 0,001320498
And 18268 was also wary lucky for you but too short to say anything. But sometime score reset do spill from block to block so it might be that... You have gained 0,000306055.
Blocks 18256, 18254, 18253, 18252, 18248 and 18247 are suspicious but less definitive. So hard to say... 18248 and 18247 are probably wrong. With them you have lost 0,00099292
And then there are smaller then 14% differences... Complicity normal with your speed and you did gain with them so total difference is only 0,001454909 in your loss(1 good block). So again total normal with one invalid if invalid is normal. But it is some time since I saw last complaint that block didn't fix itself. It was 18264 but since it is invalid that doesn't count...
EDIT:
Sorry, but Lucko could be right here - compared to the number of invalids we typically get when all is well, 2 invalids so close together is an alert - especially considering we have a recently known issue here - I just hope we don't see any more...
From that I understated the error is not constant but sporadic. So it could take some time for it to appear again. Fingers crossed...