Two different "accuracies"...why?

Question

Ok, I know it's probably not impressive to most, but I'm on my way to a 1000 rating on chess.com. I'm guessing it's because of different engine strengths or something.

I recently played the following game as white:

[FEN ""]
1. e4 c5 2. Bc4 d6 3. c3 Nc6 4. d4 cxd4 5. cxd4 Nf6 6. Nc3 Qa5 7. Bd2 Nxe4 8.
Nxe4 Nxd4 9. Bxa5 b6 10. Qxd4 bxa5 11. Bb5+ Bd7 12. Nf3 Bxb5 13. O-O-O e5 14.
Nxe5 dxe5 15. Qxe5+ Be7 16. Qxb5+ Kf8 17. Qb7 Re8 18. Qxa7 g6 19. Qxa5 Rc8+ 20.
Kd2 Kg7 21. Qe5+ f6 22. Qxe7+ Kh6 23. Qxf6 Rcd8+ 24. Ke3 Rhf8 25. Qg5+ Kg7 26.
Rxd8 h6 27. Qe7+ Rf7 28. Qe5+ Kh7 29. Nf6+ Kg7 30. Rg8# 1-0

And Chess.com estimated an accuracy rating of 94.2.

I pasted the same game into Lichess, and it estimated an accuracy rating of 99% under Stockfish 14.

So why the discrepancy in accuracies? And which is "more" accurate?

Answer 1

As you've correctly guessed, since chess engines aren't some sort of "chess gods", they wil have different opinions on a given move (just as two Grandmasters would).

I'd say there's more than that than engine evaluation though. White has a completely winning position from move 10, so it's possible that one of the accuracy calculations is "punishing" White for not finding the fastest ways to mate while the other one is just "ignoring" everything that happened after White got a queen up and made an easy conversion into a win. Think of move 12 for instance, where White loses a minor piece but still has a totally winning position.

As for which one is more correct, there's no standard definition of "accuracy" so it's a matter of opinion. In fact nobody ever talked about accuracy until chess.com started to show it in big flashy numbers at the end of games. In general I wouldn't listen to engine analysis too much when it's about positions with big imbalances. Engines are dessigned to find precise continuations and nouance in positions where humans have a hard time, not to play positions where one side is a queen up for no compensation.