Centipawn Loss Explained: What ACPL Actually Measures

You lose a game, run the analysis, and the report hands you a number: average centipawn loss 42. Your opponent finished on 38 and beat you. A friend who is four hundred points weaker posts 31 from a blitz game and wants to know why yours is worse.

The number is real and it measures something worth knowing. It also breaks in specific, predictable ways, and the players who lean on it hardest are usually the ones who have not seen it break yet.

A Centipawn Is a Hundredth of a Pawn

Stockfish scores positions on a pawn scale. An evaluation of +1.00 says White holds an advantage worth about one pawn; +0.30 is a small edge; -2.50 means Black is close to winning. A centipawn is one hundredth of that unit, so +1.50 on the eval bar is 150 centipawns.

Your loss on a single move is the gap between the engine's best move and the one you played. If the strongest continuation holds +1.50 and your move leaves +1.00, you gave away 50 centipawns. Play the engine's top choice and you lose zero.

Two properties of that definition matter more than the arithmetic.

The engine measures you against the position you actually reached. Hang a rook on move 12 and you take one large charge on move 12. Every move after that gets compared to the best available in a position where you are already down a rook, so the blunder is not billed twice.

And every move counts, including the ones that played themselves. Recaptures, checks with one legal reply, and forced trades all score close to zero loss and pull the average down. A game full of forcing sequences produces a flattering ACPL for the same reason it produces a flattering accuracy score.

The Average, and What It Hides

ACPL is the total centipawns you gave away divided by the number of moves you played. Rough bands look something like this:

ACPL	Reads as
Under 20	Master-level accuracy over a full game
20 to 40	Strong club play, few large errors
40 to 70	Solid with a handful of real mistakes
70 to 100	Blunders are deciding your games
Over 100	Multiple pieces or exchanges dropped

Treat every row as approximate. ACPL depends on which engine ran, how deep it searched, and how the tool handles opening theory, so the same game scores differently on two sites. A tool that skips your first ten book moves reports a higher ACPL than one that counts them at zero loss.

Game length distorts it too. One 400-centipawn blunder in a sixty-move endgame adds seven to your average. The same blunder in an eighteen-move miniature adds twenty-two. Your play was identical; the number moved by fifteen.

The Pawn Scale Does Not Match Winning Chances

Here is the failure that matters most, and it takes two positions to see.

In the first, you are up a rook at +9.00 and you play a slack move that leaves +8.00. You gave away 100 centipawns and you are still completely winning.

In the second, the game is level at +0.30 and your move leaves you at -0.70. Also 100 centipawns. You just handed over the game.

ACPL charges both moves the same, because a pawn is a pawn wherever it sits on the scale. Your chances of winning do not work that way. At the top of the scale you can shed material for a long time before anything changes; near equality, a third of a pawn decides the result.

Forced mates expose the same crack. Miss a mate in three and keep a +6.00 position and the pawn scale registers a modest loss, because there is no honest way to price checkmate in pawns. That miss decided the game.

Why We Grade on Win Probability

Chess It Up runs every evaluation through a conversion first. The formula, borrowed from Lichess, turns a centipawn score into the chance of winning from that position:

win% = 1 / (1 + e^(-0.00368208 × cp))

The curve is steep near 0.00 and flat once someone is winning, which is how the same 100 centipawns can be worth two different amounts. Run the two positions above through it:

• +0.30 to -0.70: win probability drops from 52.8% to 43.6%. You lost 9.2 points.
• +9.00 to +8.00: win probability drops from 96.5% to 95.0%. You lost 1.5 points.

Every grade in the review comes from that second column. The move that costs 9.2 points is a Mistake. The move that costs 1.5 points is an Excellent. The bands run like this:

Win probability lost	Grade
Under 1 point	Best
Under 3 points	Excellent
Under 5 points	Good
Under 9 points	Inaccuracy
Under 22 points	Mistake
22 points or more	Blunder

Those thresholds sit on top of extra checks for sacrifices, forced replies, opening theory, and missed opportunities, which our guide to move classifications walks through. The core idea is the one above: a move is judged by how much of the game it cost you, not by how much wood the evaluation moved.

Using ACPL Where You Find It

Lichess reports average centipawn loss on every computer analysis, and plenty of players like having one number to watch. Three rules keep it useful.

Compare it to your own past games and nothing else. Your ACPL from Tuesday against your ACPL from Monday, same site, same time control, tells you something. Your ACPL against a friend's tells you almost nothing until you know how long their games ran and who they played.

Read twenty games, not one. A single ACPL is one blunder away from meaningless. The mean across a month is a real measurement, and it responds faster than accuracy when your play changes, because it weighs mistakes by size instead of sorting them into buckets.

Then go find the moves. An ACPL of 55 tells you that something went wrong and refuses to say what. Sorting the game by the largest drops takes ten seconds and gives you three positions worth studying. Our guide to analyzing your own games covers what to do once you have them.

Frequently Asked Questions

What is a good ACPL?

For a rapid game at club level, under 40 is a good result and under 25 is strong. Blitz runs higher for everyone. Rather than aiming at a number, watch your own average fall over a few months of play.

Why is my ACPL bad when my accuracy looks fine?

The two metrics disagree by design. Accuracy sorts your moves into grades and averages the grades, so a Blunder and a catastrophic Blunder score the same. ACPL adds up the raw damage, so one disaster can wreck the average while forty clean moves keep the percentage respectable.

Do opening moves count toward it?

That depends on the tool. Some skip moves that appear in their opening book, some score every move from the first. It is one of the reasons two sites report different numbers for the same game.

Does Chess It Up show an ACPL number?

No. The free game review reports accuracy plus a count of each grade for both players, and the grades come from win probability rather than raw centipawns. You can still see the evaluation before and after any move in the review and read the drop yourself.

Will lowering my ACPL raise my rating?

Only if you lower it the honest way. Fewer hung pieces and fewer moves played on autopilot will do it. Trading into dead positions and shuffling pieces will also do it, and that habit costs rating while the number improves.

Check Yours

Take your last ten games, run them through the analyzer, and note the accuracy and blunder count on each. Ten minutes gets you a baseline that means more than any table in this article, because it is built from your games against your opponents at your time control. Come back in a month and measure again.