Fallacy of accuracy

If you ever think chess.com's accuracy score is a perfect representation of a chess performance, just look at this:

By the way, Black is Stockfish 10 running in a 5/2 minute time control, with significant hardware resources, and white is another engine running on a longer time control. So both engines are obviously much stronger than chess.com's analysis engine. So not only does white get a lower score than black, despite "dominating the whole game", but they both get lower than 99% accuracy scores, and 99% accuracy isn't that hard for even a master level human to reach. Of course, a stronger engine would have done a better job recognizing how strong white's moves were, and how appropriate black's responses were, but even so this is a reason to doubt measures like CAPS.

Here is the full game