ACPL is absolutely meaningless outside the individual game from which it is recorded, and even then, what can be read into it will be wildly different from one game to the next. I have no idea why lichess or any other service even bother giving it as a stat in insights or similar, it is not useful.
It is definitely not the case that low centipawn loss necessarily equals particularly good play.
@Chuck_Fess Ahem, low centipawn loss means that you were pretty close to making engine moves. You need to account for things like long endgames and such, but it is a useful metric. And it is certainly better than what chessdotcom uses (CAPS, which is just so bad, it's not even funny).
@Chuck_Fess ACPL alone is not a great indicator of quality in a game. And Low ACPL does not always equal great play, thats right. But calling it useless is just plain wrong. It is a very useful metric for those who know how to interpret it.
"Ahem, low centipawn loss means that you were pretty close to making engine moves."
No, it does not. It *can* be the case, but not necessarily. It can also simply mean that your opponent played terribly. Lichess' system stops adding to your total centipawn loss whenever the evaluation is above a certain number. If your opponent blunders away their queen at some point, then moves that are objectively not very good might still not add to your "score."
Also, the length of the game matters. There are two games that are identical up to move 30, with white having a huge advantage; one black player resigns at this point, while the other plays on until the inevitable checkmate on move 70. Did the first player, whose opponent resigned, play worse than the second? Of course not, that would be a ridiculous thing to think. But the second player will have a much, much lower ACPL.
It can also mean you simply play boring, unimaginative lines with little risk.
As I said, it means absolutely nothing outside the context of a single game, and even then what it can mean in a single game varies widely. Whether someone played well requires qualitative analysis - that is to say, it cannot be described in numbers alone. It requires looking at the context.
Here is a game where I posted an ACPL of 3.
lichess.org/MGvXc8gOIs the ACPL so low because I played incredibly? Or is it because for more than 75 percent of the game, it was virtually impossible for me to actually add to my ACPL calculation, because the evaluation was at greater than +10? My opponent making a horrible blunder very early in the game does not mean I did anything special. You absolutely have to look at the actual game to make any sense of the number, and thus any attempt to lump two or more games together to get some ACPL trend is pointless. Likewise, it is pointless to try to compare two separate players' performances, or one player's performance in two different games, by looking at their ACPL.
And none of this has even gotten into the flaws in the computer analysis. The depth of Lichess' server analysis is not very high.
@Chuck_Fess "flaws of Lichess computer analyses" are next to nonexistant, and absolutely nonexistant compared to the hot garbage that is your understanding of chess.
Also one weird example of a game that you played does not make ACPL a useless metric. You just don't seem to know how to use that data correctly. The questions you need to anwser are:
How cleanly was the game simplified?
How sharp was the opening ?
How long was the game?
Has the game been dragged out unnecessarily?
Was there time pressure?
If you can make an assessmet of the game correctly, you can start comparing the game to other games that would belong in the same category.
@Chillkroete77 is right in that regard, lower ACPL , means closer to engine preferred lines. It that 960 game your opponent left you no choice but the very obvious winning moves that accounts for low ACPL.
Look at moves 12, 13, and 16 according to the engine analysis of the game I posted. All three should have been inaccuracies or worse, but were not labeled as such and added nothing to the average, because of the threshold. My average did not go up, but nor did I play the “engine-preferred line”. This is because that number is not telling you what you think it is. In this case, the low number means that most of the game was spent in a state where I couldn’t have realistically increased my average.
The player I beat had an average of 51, which is not *that* high, given that he had lost the game by the third or fourth move. I wasn’t as good as the number suggests, and neither was he. His would have been higher, but he too spent most of the game unable to change his average, because of the threshold.
I would guess that as the ratings of the players get higher, so to does he likelihood of low ACPL corresponding with very good play. But for most games, that assumption does not hold. There are too many other factors that might be involved.
Calling this very straightforward truth “garbage” sure is fun, but it doesn’t convince me to give up on elementary logic and mathematical principles.
Your questions you posted, you know what those are? Those are examples of qualitative analysis, exactly the sort I am saying is necessary to make any meaningful inferences. You don’t get any answers to those just by looking at the number on insights as if all the games are the same.
I’m curious how you think telling me that my ACPL was low because my opponent played poorly supports your argument and not mine. Because that is literally exactly what I wrote when I posted the game. What are you even arguing?
The argument I have put forth in multiple posts is: low ACPL does not necessarily mean particularly good play. Your tone and insults suggest you disagree with that, but then the arguments you put forth in your comment actually support my argument.
@Chuck_Fess You seem to be a native speaker yet somehow I get the feeling you have trouble understanding some simple points I have made. I have never referred to your "truth" as garbage but to your understanding of the game, compared to Lichess server analyses, that according to you is flawed enough to be a problem to rewieving a game or sets of games. This is not meant to rag on you as a player but it is just absurd to try and pretend that a deeper server analyses would be required for anything but the highest levels of chess.
regarding your last question: It is very simple: games that your opponent throws completely at move 4 and you both do god knows what for the rest of the game are rare, and no one uses those as an example of great play or to derive any sort of information from them. They don't count and no one gives a shit.
Why you think that presenting this game means that ACPL is completely useless is beyond me. Your argument seems to hinge on your belief that players can not conduct a meaningful assessment of the game and draw conclusions from that. That is utterly wrong.
if i can give my opinion acpl is as important as any stat. Let's say pieces moves by phase. Let's say i move the queen in opening above 20% of times. More than bishop or knight. It may suggest that my opening are wrong, because i don't develop my pieces before the heavy weights. That i am running the queen from square to square as my opponent develops. Let's say loss/winning by castle type. let's say i win over 70% when king side and 10% as queen side. It may suggest i don't know how to play that kind of positions. Let's think pieces inbalances by phase. Let's think i have -0.2 in openings, 0.8 in middlegame and 3.1 in endings. It may suggest i am strong at endings, but in other game phases i can't develop properly in order to gain advantage. And this go on and on. Stats mean what means. If i eat two chickens and you eat none, in average we eat 1 each. You died hunger. I grow fat. You can never use a indicator and focus the worst example possible. I don't make my games thinking about acpl, but it is usually a good indicator to check if my game (that i know it is a normal one) was accurate or not.
If you check the link i posted about myself you have:
openings: 33.5
middlegame: 89.0
endings: 55.8
This is kind of expected. What i get is... in openings i don't know the theory as much as i should; comparing endings with middlegame, probably my endgame is not so good, as with fewer pieces and more clear best moves i have a close value compared with middlegame. Probably it should have a value close to 0 in openings (that's theory known after all) and much inferior in endings when compared with middlegame. The phases i have to improve are that two. For middlegame, i can use other indicator, for example, opportunism.
All this indicators are cool and important if we know how to use them and do not focus to much in the raw values.
my two cents