lichess.org

Very low centipawn loss?

Has the analysis tool been adjusted recently? I've noticed that my games are often very low avg. CP loss, even though they are full of very grave mistakes. Example of a 17 centipawn loss game (black pieces), where I should have been 2 clear pawns down:

lichess.org/k9YSSKqE

You might be forgetting the word "average". For half of the game, you made the best available moves (more-or-less). The fact that you were totally winning doesn't change the fact that none of those moves "lost" any centipawns.

A hypothetical scenario: If you blunder your queen on move 2, and then your opponent takes 100 moves to win, then your *average* centipawn loss might be very low (your opponent's would probably be quite high), even though you were down a queen the whole game with no compensation. Roughly speaking, this is because your queen blunder will have lost 900 centipawns, and 900 averaged over 100 moves is just 9.

it is true, but i played in chess.com until some weeks ago, my game didn't improved that much, and there i had much more imprecisions and mistakes. it seems to few failures here. or too much there, i don't know...

i'll export one game, analise there and compare. it may be the depth difference though...

@MetallicIntuition Yes the guy above hit the spot. The game is not particularly accurate, your opponent just diluted the mistakes by dragging the completely lost game out from 25 moves to 50. This is also part of the reason some bullet games seem to be more accurate on average then they are.
Here is a hypothetical scenario: after a shitty game of 25 moves (75 ACPL) both opponents shuffle their king arround for 50 moves in hopes of flagging the other resulting in no additional centipawn loss.
Now we have a 75 move game with 25 ACPL.

This is why I would love a feature that would allow us to see the ACPL for a specific portion of the game, or being able to cut away the formality of playing out a +/- 10 game to mate from the statistics.

@Morozov "This is why I would love a feature that would allow us to see the ACPL for a specific portion of the game, or being able to cut away the formality of playing out a +/- 10 game to mate from the statistics."

chess.com do that, maybe soon we have it too here

you can calculate the ACPL for any set of moves you want. Just look at the evaluation before and after a move and then you see the CPL for that move. From that you can easily derive the ACPL for any set of moves.

Reconnecting