why is chessbase analysis so different from lichess analysis?

Hi,

i wondered, what is my precision according to chessbase let's check analysis in one of my best wins, because i know i didnt cheated and i am aware that sometimes i have very low centipawn loss in some of my games. Howevere... i discovered very big discrepancy between lichess accuracy and chessbase engine correlation.

https://imgur.com/gPOVQQ7
https://imgur.com/CrB8fCx

(from my second best win since the first best win where i have better result, there is only few moves and chessbase dont give evalutation stating there is not enough moves)

So appart from Niemans drama, can someone explain to me, why there is so big difference? I know that lichess is stockfish 15 and i run chessbase with deep fritz 14, however i ran the analysis with enough time to compensate for weaker engine.)

Because of their different methods in calculating "accuracy".

Lichess, being open-source, uses centipawn to calculate accuracy.
https://lichess.org/page/accuracy

On the other hand both chessCom and chessBase, being closed source, uses some different method to do so. But, since, they are closed source, it's bit too difficult to predict their techniques accurately.

but the difference si huge. 27% seems like a poor game, 85% seems like very good game where was not much to improve. So those two companies even cant agree on rough estimate on how i played? Very good or very poorly? Is is kind of dissapointing. And there is one even more ridiculous thing: lichess says, that black played better, chessbase says, actually, the oposite - white played better somehow (even though he lost, eventually).

https://lichess.org/xDcQCep2/

I gues... you can play better (more moves like engine) and than blunder and loose anyway.. but...it is kind of confusing, really.

Thanks for the link, i will check it out.

Shame you need premium on Chessbase to analyze your games.

Never really giving 'accuracy' much thought, for me it yo-yos depending on whether I win or not.

As an example, last night I played a 15|10 rapid game, low 1600 rating. Opponent lost a pawn on move 7 and managed to turn this into a loss of a Knight. There was some compensation as I was under-developed, and the opponent had a strong attack. After defending for the next 20 moves - by no means the best defensive moves but adequate will no blunders. By move 27 I was still a Knight up and fully developed so started to swap pieces off any chance I got. Still all on the backfoot and having used 90% of my time. Game turned on a passed rook pawn that I started to push because my opponent switched tactics to flag me and consequently lost two pawns giving the Rook pawn a clear run distracting any further attacks. By no means my best game - certainly no great moves -, but happy I managed to keep things tight during the time trouble - only one significant error which would have been survivable, but my opponent didn't grab.

Result: Accuracy of 96%!, acl of 27, with one mistake (which I consider to be a wee bit dubious). My opponent scored one blunder + an inaccuracy and 81%, acl 47 (lowered because of the flagging attempt).

The accuracy formula needs to factor in some sort of position complexity measure to be truly useful in terms of identifying high-quality games.

Chess.com's accuracy measure is a bit more solid but suffers from the same problem, a relatively simple games can yield very high values.

@segomo said in #1:

Hi,
So appart from Niemans drama, can someone explain to me, why there is so big difference? I know that lichess is stockfish 15 and i run chessbase with deep fritz 14, however i ran the analysis with enough time to compensate for weaker engine.)

because everyone has they own definition of accuracy, blunder, mistake, inaccuracy. Hence it unreasonable to assume that you would get same moves marked inaccuracies or get same numerical value for accuracy. chess base accuracy might be just number of moves matching engine selection. which obviously gives differeren valuation from lichess formula