Accuracy seems low compared to graph game domination • page 1/1 • Lichess Feedback • lichess.org

If the graph shows dominance for white, how can the accuracy of the game be at 20% for white?
The domination of the game was at 4% for me and I end up winning after White resigned.

Is the accuracy of a game affected by one bad phase of the game?
If so, then accuracy should be shown in three phases.

Below is only part of the analysis results from Lucas's chess analysis, using Stockfish 15 (30 secs + depth 28).
Domination: 42.00% White - Black 4.00%
Opening: 3282 White - Black 3235
Middle game: 2619 White - Black 2321
End game: 438 White - Black 3271

https://lichess.org/7WcncRTp/

Human moves vs Stockfish best engine move compared.
Identical = 100%
Move#. White - Black

100% - 092%
100% - 096%
095% - 067%
100% - 041%
038% - 030%
100% - 100%
000% - 100%
100% - 099%
100% - 060%
095% - 100%
100% - 100%
100% - 100%
100% - 022%
100% - 100%
100% - 049%
100% - 000%
100% - 100%
100% - 000%
100% - 007%
100% - 100%
100% - 100%
100% - 100%
100% - 064%
100% - 100%
000% - 100%

Total white matching best moves 20/25, but Lichess Accuracy gives 20%
Total black matching best moves 12/25, but Lichess Accuracy gives 85%

There were a lot of best moves played for an accuracy that doesn't seem to correlate.

If the graph shows dominance for white, how can the accuracy of the game be at 20% for white? The domination of the game was at 4% for me and I end up winning after White resigned. Is the accuracy of a game affected by one bad phase of the game? If so, then accuracy should be shown in three phases. Below is only part of the analysis results from Lucas's chess analysis, using Stockfish 15 (30 secs + depth 28). Domination: 42.00% White - Black 4.00% Opening: 3282 White - Black 3235 Middle game: 2619 White - Black 2321 End game: 438 White - Black 3271 https://lichess.org/7WcncRTp/ Human moves vs Stockfish best engine move compared. Identical = 100% Move#. White - Black 01. 100% - 092% 02. 100% - 096% 03. 095% - 067% 04. 100% - 041% 05. 038% - 030% 06. 100% - 100% 07. 000% - 100% 08. 100% - 099% 09. 100% - 060% 10. 095% - 100% 11. 100% - 100% 12. 100% - 100% 13. 100% - 022% 14. 100% - 100% 15. 100% - 049% 16. 100% - 000% 17. 100% - 100% 18. 100% - 000% 19. 100% - 007% 20. 100% - 100% 21. 100% - 100% 22. 100% - 100% 23. 100% - 064% 24. 100% - 100% 25. 000% - 100% Total white matching best moves 20/25, but Lichess Accuracy gives 20% Total black matching best moves 12/25, but Lichess Accuracy gives 85% There were a lot of best moves played for an accuracy that doesn't seem to correlate.

glbert

#2

lichess accuracy has little to do with matching best moves: https://lichess.org/page/accuracy

Toscani

edited

#3

An accuracy of 20% means you only played terrible moves, which is not true according to the above proof I have shown.
An accuracy of 100% means you played all the preferred Stockfish moves.

I did exactly that, proving that the accuracy needs to be shown in phases. The 0% moves are affecting the overall accuracy results. That single bad move hurts the true accuracy of the overall game.

Accuracy 20% represents how much white deviated from the best moves. Well, he deviated only 5 times.
Black deviated 13 times and I get 85% accuracy?

An accuracy of 20% means you only played terrible moves, which is not true according to the above proof I have shown. An accuracy of 100% means you played all the preferred Stockfish moves. I did exactly that, proving that the accuracy needs to be shown in phases. The 0% moves are affecting the overall accuracy results. That single bad move hurts the true accuracy of the overall game. Accuracy 20% represents how much white deviated from the best moves. Well, he deviated only 5 times. Black deviated 13 times and I get 85% accuracy?

Toscani

edited

#4

I have removed the last final move before resigning. It should have never been included in the overall accuracy of the game. That last blunder could in theory have been done to distort the accuracy of the calculations. Hiding the facts of all those best moves. Here is a closer truth of the accuracy of the overall game, just before resigning to that accidental oversite.

White
0 inaccuracies
0 mistakes
1 blunder
25 Average centipawn loss
89% Accuracy

Black
4 inaccuracies
1 mistake
1 blunder
47 Average centipawn loss
84% Accuracy

Feature request: Show accuracy by phase. (Opening accuracy, middle game accuracy, end game accuracy)

https://lichess.org/study/0XNw02mf/

Thank you for your hard work in creating an accuracy check of the chess games.

I have removed the last final move before resigning. It should have never been included in the overall accuracy of the game. That last blunder could in theory have been done to distort the accuracy of the calculations. Hiding the facts of all those best moves. Here is a closer truth of the accuracy of the overall game, just before resigning to that accidental oversite. White 0 inaccuracies 0 mistakes 1 blunder 25 Average centipawn loss 89% Accuracy Black 4 inaccuracies 1 mistake 1 blunder 47 Average centipawn loss 84% Accuracy Feature request: Show accuracy by phase. (Opening accuracy, middle game accuracy, end game accuracy) https://lichess.org/study/0XNw02mf/ Thank you for your hard work in creating an accuracy check of the chess games.

Toscani

#5

Feature request: Can the accuracy be shown when we press on a move or directly on the graph.
Example: If I press move 46, the accuracy does not change, but what I would like to see is the accuracy as the game progresses.
The fact that the person had an oversight or did it on purpose to blunder or do a weaker move affects the true accuracy of the game line. If these games get compiled by accuracy per move, the mainlines should stand out. Then the odds of playing those main lines by rating can be plotted for cheat detection purposes or to solve a mainline.

https://lichess.org/A01E8xWR/black#46

Feature request: Can the accuracy be shown when we press on a move or directly on the graph. Example: If I press move 46, the accuracy does not change, but what I would like to see is the accuracy as the game progresses. The fact that the person had an oversight or did it on purpose to blunder or do a weaker move affects the true accuracy of the game line. If these games get compiled by accuracy per move, the mainlines should stand out. Then the odds of playing those main lines by rating can be plotted for cheat detection purposes or to solve a mainline. https://lichess.org/A01E8xWR/black#46

borninthesixties

#6

The "accuracy" numbers provided by lichess are sometimes quite different than those provided by chess.com. I played a few games over there yesterday, then imported and analyzed them here. In one game lichess gave me 48% accuracy while chess.com gave me 69.7%. In another game, lichess gave me 91% while chess.com gave me 80.1%. In a third game the numbers were closer (78.0 lichess vs 72.6 in chess.com) Not sure if this is really a useful number based on those discrepancies. Maybe they haven't worked out all the bugs.

glbert

#7

Not sure if this is really a useful number based on those discrepancies.

there is really no reason why the lichess number should agree with the chess.com number. both of them are called "accuracy", but neither of them is accuracy in the usual sense of the word – https://en.wikipedia.org/wiki/Accuracy_and_precision
basically they just picked a name that sounds good.

> Not sure if this is really a useful number based on those discrepancies. there is really no reason why the lichess number should agree with the chess.com number. both of them are called "accuracy", but neither of them is accuracy in the usual sense of the word – https://en.wikipedia.org/wiki/Accuracy_and_precision basically they just picked a name that sounds good.