How to Estimate your ELO for a game, using ACPL, and what it realistically means?

Average Centipawn loss.
Blitz 75.40
I am 2010 at blitz. Im not sure if your chart is accurate.

and as mentioned in this thread before ACPL in single game is function of players skill and complexity of the game.

I've see IM friend mine to play below 5 acpl and over 30 acpl. If situation gets complex enough no one is gonna fine best or almost best moves

@Alientcp said in #11:

Average Centipawn loss.
Blitz 75.40
I am 2010 at blitz. Im not sure if your chart is accurate.

Given a (golden ratio - 1)% on leeway on your ACPL - that would be towards the top range of expected ELO.
According to this, your ELO could be between to 915 and 1945 ELO, on average, given leeway.
Whereas, others in the thread were towards the bottom of the range, given a ~1600 ELO game.
It would seem to me, that both examples busted through the top and bottom end of the leeway.

I wonder, is the correlation actually inverse, or should there be an absurd amount of leeway to prove a weak correlation?

I wonder, would this be the case, if only games that were Checkmate were included in this, because games that don't end in checkmate have numbers that would have expanded one way or another as the game finishes.

I believe that this ACPL method of evaluation is worth something - but it would appear from the examples given in this thread that, that something its worth is either worth negative evaluation to humans, or a weak correlation with an absurd amount of leeway.

Because its been busted on both the top side and the bottom side, using 2 counter-examples now from players. What does this actually mean, you think?

Maybe, one method of evaluation, ACPL, given computers, strengthens their gameplay, but doesn't strengthen human players gameplay, overall, and when it does its on a weak basis?

@Approximation said in #1:

Does that mean that the highest rating possible is 3100?
or does negative acpl exists

Using ACPL for rating estimations is simply bonkers. CPL varies a lot depending on the openings/positions one likes to play. Also stuff like playing clearly drawn endgames or agreeing to a draw has an influence. CPL is also influenced by the strength of the opponent.

@Kewin3 said in #14:

Does that mean that the highest rating possible is 3100?
or does negative acpl exists

Yes, In theory, by this method of approximate calculation, a zero ACPL loss in human play would give a Human chess rating of 3100; which in reality Magnus Carlsen has a rating close to 2900, actually 2882, and his most accurate chess game was 6.62 ACPL, where this method does well to approximate his game/rating at this level. Maybe, ELO per inverse ACPL is slightly over-estimated, assuming that is his top game he played, according to engines that calculate a game, measuring it with ACPL. Actually I think the data point should go through 2882 and his average ACPL, instead of his best ACPL game.

Now, what does this mean for Computer Chess engines that play > 3100 ELO? It means the ACPL method of evaluation breaks down, where it either needs to be supplimented by other chess logic to do better, or simply throw this method of calculation out the window.

So, is the best chess possible games played with a zero ACPL method strategy, result in a rating of 3100 ELO? Maybe! Therefore, ACPL might be correlated, and noteworthy to this point, but worth less and less, beyond it. Also, perhaps, logically, some people have found ways to think where ACPL already isn't worth much in their assesment of a chess position - idk. That could result in the weak correlation, or lot of leeway that is necessary to make the model work.

And, I don't know the answer to the negative ACPL question. But, in this model, a negative ACPL wouldn't have meaning, I'd think.

That equation needs to be adjusted for blitz vs classical. The numbers seem accurate for my rapid games.

Have they used this to find the best chess player in history?

Also, why 3100 instead of 3600?

The reason for 3100 instead of 3600 is 3100 is for mass calculation. 3600 would take too much calculation time. I doubt it would be more accurate since we average over several games, though for world champions we might want to use more computing time.

@Chesserroo2 said in #17:

Have they used this to find the best chess player in history?

Aside from Carlsen's 6.62 ACPL chess game - idk if another game anyone has ever had at the top level has been more accurate than that, according to ACPL. And, as far as History goes, being that chess players from different Eras realistically don't compete against each other, and even though idk the answer - I'd speculate, that it would be hard to find a player who's more accurate than Carlsen, on an ACPL basis over many games, although finding a single game is probably more possible, I'd think.