Its proven that AI plays better than human if AI plays moves which it actually chooses. (And if situation is not too unusual.)
But can we really trust score estimation of random move? (move which AI didn’t even planned to analyse) Especially if you have too few playouts.
there are big bugs , maybe there are also small bugs
So, in AI review if your move shows bigger score than blue move (move that AI would choose), I think its more safe to play blue move next time in similar situation instead of playing your move again. Maybe your move is better, but can buggy AI with few playouts be proof of that? I suggest to ignore the score, ignore the %, and just to look at which move is blue.