Go back
Accuracy % from computer analysis for one's games

Accuracy % from computer analysis for one's games

Only Chess

Paul Leggett
Chess Librarian

The Stacks

Joined
21 Aug 09
Moves
114051
Clock
31 Oct 20
1 edit

Hi all,

Yesterday I won a game against the number 1 seed in a club tournament on Chess.com, and I clicked the button for analysis on the site. It is some iteration of Stockfish, and they claim 24 ply.

It said White was 91% accurate, and that I (as Black) was 93% accurate.

I didn't really believe it, so I used Chessbase to do a quick check. Using the tactical analysis feature, Chessbase said that White was 53% accurate, and that Black was 55% accurate.

Clearly there are different parameters at work, but the bottom line is that I think Chessbase's figures are far more accurate.

Those ridiculously high chess.com figures only pour gasoline on the fire of those players who think that, when they lose, the other guy MUST be using an engine.

Indeed, if I did not know better, and know enough to verify before trusting, I would look at those chess.com numbers and consider both players to be suspect- and I'm one of those players, obviously. I think they are doing a disservice from good intentions.

greenpawn34

e4

Joined
06 May 08
Moves
43363
Clock
31 Oct 20
1 edit

Hi Paul,

Over at the ECF forum there is a very long discussion on players getting banned
apparently at random at various chess sites. (I think it's a misguided effort to
convince would be cheats the site owners there are on the ball.)

They banned the wrong guy and he fought back. He has been keeping
us updated. I found this email of his to chess.com quite hilarious.

Dear Chess.com Support

Thank you for your mail, one of five I have recently received from you.

You emailed me yesterday to inform me that my account was being closed
because I had violated your Fair Play Policy. I emailed back to say that as
I had not played on your site in thirteen months, I did not see how that was possible.

Today I received four more emails, within a few minutes of one another.
The first of these was the same as the earlier one. The second informed me
that I had been awarded Premium Membership. The third informed me that my
payment for that membership had been processed.

The fourth, again, informed me that my account had been closed.

Yours....

Being awarded a premier membership, taking money for it and closing down
the account! Sounds like the site is being run by that infamous Nigerian Prince.

D
Losing the Thread

Quarantined World

Joined
27 Oct 04
Moves
87415
Clock
31 Oct 20

@paul-leggett said
Hi all,

Yesterday I won a game against the number 1 seed in a club tournament on Chess.com, and I clicked the button for analysis on the site. It is some iteration of Stockfish, and they claim 24 ply.

It said White was 91% accurate, and that I (as Black) was 93% accurate.

I didn't really believe it, so I used Chessbase to do a quick check. Using the tactical a ...[text shortened]... and I'm one of those players, obviously. I think they are doing a disservice from good intentions.
At a guess Chessbase is comparing with engine first choice, whereas chess.com have some threshold for number of centipawns lost for accuracy. So 93% of your moves are within some threshold of the best move, as judged by Stockfish.

Any idea if they use an openings book? I can rattle off 10+ moves of theory in some openings and will be 100% accurate for those 10 moves. Also some endings like K+B+N vs K takes 20 odd moves from when you have the king trapped in the wrong corner to when one delivers checkmate in the right one. Anyone who can do the checkmate has a high chance of a 100% matchup with Stockfish during that procedure, since there isn't much scope for variation. It depends exactly what one does with the bishop when preventing the king breakout. So the accuracy score will be affected by this.

Paul Leggett
Chess Librarian

The Stacks

Joined
21 Aug 09
Moves
114051
Clock
31 Oct 20

@deepthought said
At a guess Chessbase is comparing with engine first choice, whereas chess.com have some threshold for number of centipawns lost for accuracy. So 93% of your moves are within some threshold of the best move, as judged by Stockfish.

Any idea if they use an openings book? I can rattle off 10+ moves of theory in some openings and will be 100% accurate for those 10 moves. ...[text shortened]... with the bishop when preventing the king breakout. So the accuracy score will be affected by this.
I haven't had a chance to deep-dive on it yet.

I suspect that you are right. Chessbase must use a "sniper rifle" approach to accuracy, whereas Chess.com thinks of accuracy in terms of the blast radius of a grenade.

s
Fast and Curious

slatington, pa, usa

Joined
28 Dec 04
Moves
53321
Clock
01 Nov 20
Vote Up
Vote Down

@Paul-Leggett
So how about averaging them out, makes that you making about 70% accurate....

mchill
Cryptic

Behind the scenes

Joined
27 Jun 16
Moves
3283
Clock
01 Nov 20
Vote Up
Vote Down

@paul-leggett said
Hi all,

Yesterday I won a game against the number 1 seed in a club tournament on Chess.com, and I clicked the button for analysis on the site. It is some iteration of Stockfish, and they claim 24 ply.

It said White was 91% accurate, and that I (as Black) was 93% accurate.

I didn't really believe it, so I used Chessbase to do a quick check. Using the tactical a ...[text shortened]... and I'm one of those players, obviously. I think they are doing a disservice from good intentions.
Slightly off topic here I found the game analysis feature more far more useful on chess.com than the accuracy %. This has been a wonderful educational tool. I see what you mean though, it had several of my completed games rated in the low 90's and high 80's (No chance IMO!)

Ragwort
Senecio Jacobaea

Yorkshire

Joined
04 Jul 09
Moves
189452
Clock
02 Nov 20
1 edit

@Paul-Leggett

https://support.chess.com/article/1135-what-is-accuracy-in-analysis-how-is-it-measured

According to IM Danny Rensch you got an A!

Ponderable
chemist

Linkenheim

Joined
22 Apr 05
Moves
669859
Clock
02 Nov 20
2 edits
Vote Up
Vote Down

@ragwort said
@Paul-Leggett

https://support.chess.com/article/1135-what-is-accuracy-in-analysis-how-is-it-measured

According to IM Danny Rensch you got an A!
Who would alert that post and why?

Okay I just got an auto-alert for this post…
...and another one for the edit...so it probably reads 3 now 😲
And I saw that ragwort has edited also, so my Question is answered About the who

Paul Leggett
Chess Librarian

The Stacks

Joined
21 Aug 09
Moves
114051
Clock
03 Nov 20
1 edit

@ragwort said
@Paul-Leggett

https://support.chess.com/article/1135-what-is-accuracy-in-analysis-how-is-it-measured

According to IM Danny Rensch you got an A!
Thanks for this! It is nice to know that it is completely different from, and has nothing to do with, the cheat detection process.

Since it is based partly on the accuracy of a response to an opponent's threats, it has the counterintuitive result of making it easier to be accurate in sharper games, since the threats are more concrete. Positional threats where there are several superficially reasonable moves could be much more of a challenge, I suspect.

jb70
State of Confusion

Lancashire

Joined
04 May 08
Moves
842388
Clock
04 Nov 20

This game gets white 99% and Black gets 19.7%


Shows the high score is easy to achieve against weak moves, indicating opponent's strength is highly significant so not a useful tool for cheat detection.

Paul Leggett
Chess Librarian

The Stacks

Joined
21 Aug 09
Moves
114051
Clock
07 Nov 20
Vote Up
Vote Down

@jb70 said
This game gets white 99% and Black gets 19.7%
[pgn]1.d4 e5 2.dxe5 f6 3.e4 fxe5 4.Qh5+ g6 5.Qxe5+ Qe7 6.Qxh8 Qxe4+ 7.Ne2 Bb4+ 8.Bd2 Qxc2 9.Bxb4 1-0 [/pgn]
Shows the high score is easy to achieve against weak moves, indicating opponent's strength is highly significant so not a useful tool for cheat detection.
Nicely done. Thanks!

Cookies help us deliver our Services. By using our Services or clicking I agree, you agree to our use of cookies. Learn More.