Free online Chess server. Play Chess now in a clean interface. No registration, no ads, no plugin required. Play Chess with the computer, friends or random opponents.
Create a gameTournamentSimultaneous exhibitions
Chess basicsPuzzlesPracticeCoordinatesStudyCoaches
Lichess TVCurrent gamesStreamersBroadcasts (beta)Video library
PlayersTeamsForumQuestions & Answers
Analysis boardOpening explorerBoard editorImport gameAdvanced search
Sign in

What is "Average centipawn loss"?

Asked by thibault
Tags analysis centipawn acpl
Activity Viewed 168135 times, last updated
The analysis page displays "36 Average centipawn loss" on the bottom right:

What does it mean?
receipt commented :
Fantastic analysis! It would be great if the program made some specific suggestions about what to study to improve too.
rioderaca commented :
Hello, how much centipawn loss ensues from a "Lost forced checkmate sequence." situation?

More explanation if needed: you had checkmate in a few moves, but blundered it - how much centipawn does lichess consider you lost there, as the evaluation was #x (mate in x) and becomes a score after you missed the sequence.

EMLEML commented :
What does average centipawn loss mean in Antichess?
MichaelBuaiz commented :
Yes, this is what I am looking for ...a way to see what I need to do to improve
8 Answers
Answered by thibault
The centipawn is the unit of measure used in chess as measure of the advantage. A centipawn is equal to 1/100 of a pawn. Therefore 100 centipawns = 1 pawn. These values play no formal role in the game but are useful to players, and essentials in computer chess, in order to evaluate positions. --

A perfect move will lose zero centipawn, but lesser moves will result in a deterioration of the position, measured in centipawn.

This value can be used as an indicator of the quality of play. The less centipawns one loses per move, the stronger the play.

These numbers are provided by the computer analysis, which is powered by Stockfish.
Chesslife commented :
This web is fantastic! Thanks very much!
PINke commented :
thank you very much =)
Mikhail04 commented :
thanks for the info
Frapator commented :
Would Lichess be able to calculate the correlation between the average player's average centipawn loss and the Glicko2 ?
jeweler commented :
Thanks a lot...the site is great!
Ritesh1304 commented :
I am so astonished and thankful to lichess....Brilliant work.....Best
seanysean commented :
thanks for the answer
Freddy47 commented :
Thanks:makes sense now
Solal35 commented :
But, why if we do a game without inaccuracies, without mistakes and without blunders we already loss centipawn ?
seanysean commented :
you should ask by q&a method of asking.
chuckwife commented :
but,i still don't kown how stockfish calculate the centipawn loss
Freddy47 commented :
Everyone: post your best(lowest) centipawn loss here!!!!

Mine is really five...I got a one but all I did was scholar's mate
skarby commented :
Like centimeter
1/100 of a meter
syyhkyrotta commented :
Thank you very much! Helped a lot.
Milenko commented :
Wonderfull website. Thaks a lot guys!
intelin commented :
Thanks this makes sense now
kaynight commented :
what is a good average centipawn loss?
And why can't the position get better - would that be a negative centipawn?
Uhohspaghettio commented :
This answer has problems. There's no such thing as a "perfect move". Computer engines have differences of opinions, particularly if they're given scant amount of time to analyze as is the case on this site. It would be nonsense to think that one opening is better than another because of how the computer has analyzed it. Sometimes one type of position may suit a particular player, while another player would do very badly in it.
nchebykin commented :
When you have no inaccuracies, mistakes or blunders but still have centipawn loss - it's because inaccuracy starts where you drop 50 centipawn loss. If you lose less on each move - you basically played without inaccuracies, but with <50 centipawn loss.
linkichei commented :
Thanks for the info!
pentille commented :
Now get it. Thank you.
che55knight66 commented :
That is so cool! THANKS!!!
Loeksnokes commented :
I would love to see a graph of average centipawn loss versus rating for each game phase (opening, middlegame, endgame, and perhaps, the transitions).
tgf commented :
I'm puzzled. How can the number always be positive ? Usually the winner's position will improve throughout the game so I would expect the winner's "average centipawn loss" to be negative. Why is this not the case ?
Arsalan-Khan commented :
Thank you. But is the 100=1 pawn just an arbitrary rate?
kart0ffelsalaat commented :
@tgf You cannot improve the position with your move due to the way the engine evaluates it. If your opponent drops his queen, and now you can take it, it will already be evaluated as brilliant for you before you actually do the move and capture. Just if you don't take, your position will get much worse.

Just try it out, analyse a game, click through the moves and watch the position value. It will never get better for a player after his own turn (or if it does, that's just because Stockfish didn't have enough time to calculate). If you have a brilliant, game changing move, the computer will already consider the position winning before your move -- all you have to do is play it.
ThantLwinoo commented :
Thank alot, for your kindness.
chris2007 commented :
The centipawn loss is the difference in the evaluation in centipawns from your move and the best move in the position. Since you cannot play a move that is better than the best move, the centipawn loss cannot be negative.
Calico992 commented :
The challenge with centipawn would seem that the computer couldn't access brilliant sacrifices (i.e. Fischer's Queen Sac Vs. Byrne) Am I incorrect?
ScorpianFire7 commented : How does the Atomic average centipawn loss work?
ScaccoLento commented :
thanks, great site, great info
Goke09 commented :
Finally, Learned what centipawn is
DrDank1984 commented :
Great breakdown. I am now able to evaluate my games more effectively.
Answered by Helvegen
So does it indicate the average centipawn loss you've made per move considering the total loss during the game?
Whatzup109 commented :
Thanks so much now I get it
OmegaDoom commented :
I'm also wondering what average means. I have 39 average centipawn loss. Does it per move or per game?
thibault commented :
what is the difference?
chessplayer_909 commented :
I think it should be per move.
chessplayer_909 commented :
I mean it is per move
RealKool commented :
@ OmegaDoom
Here is an example of average. We will calculate the average centipawn loss of white player.
1. e4 e5
2. Nf3 Nc6
3. a3
{engine analysis of a3: +10cp}
{engine analysis of its best move Bb5: +20cp}
Here the cp = centipawn.
Centipawn Loss = 20 - 10 = 10
3... a6
4. Be2
{engine analysis of Be2: +1cp}
{engine analysis of its best move Bxc6: +15cp}
Centipawn Loss = 15 - 1 = 14

We will stop here since this is just an illustration.

Average centipawn loss = (10 + 14) / 2 = 24/2 = 12
Average centipawn loss = 12

10 is from move 3.
14 is from move 4.
2 is because there are only 2 moves being considered.

If we consider 20 moves for white for example then
Average centipawn loss = (L1+L2+L3+L4+ ... L20) / 20

This centipawn loss is per game.

To get the average centipawn loss of a player, collect some games, get the centipawn loss per game then calculate it this way.
G1 = 60 centipawn loss
G2 = 25
G3 = 50
G4 = 200
G5 = 125
G6 = 42
G7 = 75
G8 = 175
G9 = 10
G10 = 88

Player average centipawn loss in 10 games = 60 + 25 + 50 + ... 88 = 850/10 = 85

Kasparov has an average centipawn loss of around 13 based on Crafty engine. This could be higher if stockfish is used since Stockfish is stronger than Crafty,
zcnsr commented :
@RealKool thank you so much for explaination!
zcnsr commented :
I have tried calculate by ur guid. and I get another results...
RealKool commented :
Which games do you use to calculate? And what other data do you use as a base.
When you compare 2 data, you have a base data and the calculated data.

With the implemented chess insights this average CPL can now be easily retrieved.
likeawizard commented :
@RealKool, I am pretty sure your average centipawn loss of a player calculation is incorrect. Your games G1 through G10 all will have different move numbers so to get the correct average centipawn loss you need to consider that. Basically to get the average centipawn loss of a player you would have to sum up the CPL of every move of every game and then divide by the total number of moves over all games, which is equivalent of taking a weighted average of the games by move number which would look something like:

(G1 * m1 + G2 * m2 + .... + G10 * m10)/(m1 + m2 + ... + m3) where m1, m2... is the number of moves made in that game.

(which really is the same as counting all centipawn loss over all games and dividing by total moves of all since G1 * m1 is the average times the number so it gives the total centipawn loss of game 1)
cd2017 commented :
@RealKool I don't understand about your explantion
InnateAluminum commented :
@cd2017 Which part of RealKool's explanation is confusing?
Answered by bizdev
great site! and good explanation about centipawn loss. As another writer has said, it would be useful to know the common themes/patterns around my centipawn loss. At the moment the analysis will give it me by piece, which is really good, but I would like it across some kind of thematic.

i.e. centipawn loss in the following categories:

missed a knight fork, missed a pin, missed a discovered check, missed a double attack, missed a skewer.

Something like this would be much more helpful to me to improve my game - I think.
e4ia commented :
Thanks for info, as others above! I've never heard of this term/measuring "concept" before, and I've been playing chess for + 25 years! It will be interesting (for me), to see what % of the games I lose, for example, when my centipawn total is 25% + of my opponents'.....then at 50% or higher, how the losing (winning) % of games increases/decreases....what the average is in drawn games etc.
Answered by kettwiesel
More infos about average centipawn loss (acpl):
Answered by JohnMiller

I'd like to throw an "old hat" of L. Pachmann's into the ring: he suggested evaluating a position not only by the material balance, but also taking into account some very simple statistics of a position
1. Count your number of "free" squares, it is those, where you can move a piece of yours without risk of it being capture, despite of it's value
2. Repeat this for the free squares of your opponent
3. Count your number of legal moves
4. Repeat this for your opponents legal moves
In his opinion material is only one out of four factors influencing the game, the others being time, space and initiative. Steps one and two try to account for space, while number three and four do so for space and time.

Best of regards
JohnMiller commented :
I might add, that these have to be taken into account in addition to the material factors, of course :-)
Francesco_Super commented :
Thanks for the information, John Miller :)
billywm commented :
The engine evaluation of a position encompasses all of this, and more. It assigns centipawn values to positional advantages such as space, advanced passed pawns, king safety, "good bishop/bad bishop", etc. This is all represented in the final number it spits out. Being +3 with equal material means that your positional advantages are worth 300 centipawns (e.g. an extra piece or three extra pawns)
Danko_25 commented :
Luděk Pachman's books are still good for intermediate players. Tempos/space/material tradeoffs are good concepts to better understand the game, especially openings. Pachman was teaching to better understand the quality of pieces from quantitative elements. Mid-game space control is probably one of strongest engine features, I believe mid-game zugzwang theme happens more against computers than skilled humans.
Answered by Razer531
Centi-pawn. Centi-meter=1/100 of meter
centipawn=1/100 of a pawn.
Therefore, if your centipawn loss is 23, then that means that, on some strategical and positional level, you've made concessions that are equivalent to losing 23% of a pawn.
Answered by DontReadThis
I wish someone would make a way that you could estimate the playing strenght of the player.Like it would say your estimated game performance is 2150.It would maybe use an average centipawn loss scale and other factors.

It would be nice to know if you played higher than your rating
billybilly71 commented :
If you play tournaments here you get a rating performance after based on how well you performed v the opposition which is useful.
Answered by Paradise_Pete
A simple answer:

Stockfish gives each position a numerical score. for each move, it gives the position resulting from your move a score and then subtracts it from the score it thinks is the best possible one.

It adds up those differences and then divides by the number of moves.

Only registered members with one week of lichess activity can contribute to the Q&A.