Centipawn Loss Explained for Beginners

A Centipawn Is 1/100th of a Pawn. That's the Whole Idea.

A centipawn is one hundredth of a pawn. It is the unit chess engines use to score how good a position is. If an engine says +1.50, White is ahead by about a pawn and a half. If it says -2.10, Black is ahead by about two pawns' worth of position.

That minus sign matters. By convention, the number is always given from White's side. Positive means White is doing better. Negative means Black is doing better. Zero means the position is roughly equal.

Where did this unit come from? Chess players already agreed on rough piece values, in pawns. A knight or bishop is worth about 3 pawns. A rook is worth about 5. A queen is worth about 9. Engine builders took that same scale. They split each pawn into 100 pieces. It works the way a dollar splits into 100 cents. That gives the engine room to say a position is worth 1.35 pawns. It does not have to round to 1 or 2. A centipawn is just that: a pawn, counted in cents.

There is one big exception. Sometimes an engine spots a forced checkmate. That means a string of moves leads to checkmate no matter what the other side tries. When that happens, the engine stops using centipawns at all. Instead it shows a mate score, something like "M5" or "#-3." M5 means mate in 5 moves for the side to move. No pile of extra pawns compares to checkmate. So the engine switches units once mate is guaranteed. That is the whole idea: a pawn, cut into 100 pieces, counted from White's side. Unless mate is coming. Then pawns stop mattering at all.

Average Centipawn Loss Adds Every Move's Cost Into One Number

Centipawn loss, or CPL, is the gap between two evaluations. First, the engine checks the position before your move. It finds its own best move there. Then you make your move. The engine checks the new position. The difference between those two scores, in centipawns, is your CPL for that one move.

Play the engine's top choice every time, and your CPL for that move is zero. You cost yourself nothing. Play something worse, and the gap shows up as a positive number, no matter how small.

Average centipawn loss, spelled out fully, means the average of your centipawn loss across an entire game. People shorten it to ACPL. Add up the CPL from every move you made. Divide by the number of moves. That single number is your ACPL for the game. If you played the engine's best move every time, your ACPL would be zero too. Nobody actually does that. Real games always have some loss in them, even from strong players.

This is the plain definition. Sit with it before you look at trend lines. Sit with it before you compare your ACPL across a season of games. One game gives you one number. That number is how far off the engine's ideal path your moves drifted, on average, from the first move to the last. It does not care why you drifted. Maybe you missed a fork. A fork is one move that attacks two of your pieces at once. You can only save one of them. Maybe you left a piece hanging. That means you left it undefended, so it could be captured for free. Maybe you just misjudged an endgame. An endgame is the late part of a game, when few pieces are left. The number does not say which one it was. It only counts the gap between what you played and what the engine would have played. Then it reports the average.

Inaccuracy, Mistake, Blunder: These Labels Are Not Fixed Centipawn Cutoffs

When an engine tags a move, it often calls it an inaccuracy, a mistake, or a blunder. A blunder is a move that throws away a big part of your position at once. Many players assume each label is a fixed centipawn range. It is not that simple. Lichess, the free chess site, does not use a raw centipawn cutoff at all. It looks at how far your move drops your winning chances. Winning chances just means the odds you were on track to win before you moved, out of 100. A drop of about 10 out of 100 is an inaccuracy. A drop of about 20 is a mistake. A drop of about 30 or more is a blunder. Chess.com uses its own system and does not share its exact cutoffs. So there is no single centipawn number that marks a blunder everywhere. The label depends on the tool you use. This is one more reason the number alone cannot tell you what you keep doing wrong.

The Number Tells You How Much You Slipped, Not Which Mistake You Keep Making

Here is the scale, so the number means something. Using ChessBase's Centipawn Analysis tool, strong grandmasters playing long time-control games typically average 15 to 25 centipawn loss per game. Club players typically average 30 to 60. That gap holds up in more than one source. One rough public estimate puts a grandmaster's average under 20. It puts a brand-new player at around 150. In the 12 classical games of the 2018 World Chess Championship, Magnus Carlsen's average centipawn loss came out in the single digits.

Time control matters too. Blitz games give each side only a few minutes to think. They tend to push the average higher than slower games, for players of every skill level. So picture two games. One is 3-minute blitz. One is 30-minute rapid. Both show a score of 40. They are not really telling you the same thing about your play.

Here is the honest part. None of these numbers tell you what you actually did wrong. They tell you how far off the engine's best path you drifted, on average, across a whole game. A grandmaster averaging 18 and a club player averaging 45 both know they made mistakes. Neither one knows this from the number alone. Was the mistake a fork they walked into? Was it a piece they left hanging? Or was it a pawn structure they let rot for twenty moves? The number does not say. This kind of average is a scoreboard. It shows the final score. It does not point back at the specific play that cost you the game.

Two Games With the Same Score Can Have Completely Different Problems

Here is a made-up example. It is not a real reported result. It is just a way to picture the problem clearly. Say two different players each finish a game with an average centipawn loss of 45.

Player one played almost the whole game close to the engine's line. Then, on one move near the end, their king had no escape square. A rook slid down the open file and delivered mate on the back row. That is called a back-rank mate. Your back rank is the row of squares closest to you. Your king was trapped there, behind its own pawns. It had no escape square. A single rook or queen can end the game right there. That one move loss was so large it dragged the whole game's average up to 45, even though every other move was close to perfect.

Player two never blundered anything that big. Instead, they missed five smaller chances across the middlegame. The middlegame is the stretch of a game after both sides finish moving their pieces off their starting squares. A knight fork here, a pawn push there, a tactic that would have won material but got overlooked each time. None of those five moves were disasters on their own. Added together and averaged, they land at the same 45.

Same number. Different games. Player one has one bad habit to fix: check for king safety before the position gets sharp. Player two has a different habit: look harder for tactics in the middlegame before moving on. That average of 45 cannot tell you which player you are. You would have to go back through every move yourself to find out, game after game. That is exactly the part single-game review makes so slow.

Find the Recurring Pattern Behind the Number, Not Just the Number Itself

BlackBox Chess does not show you an average centipawn score at all. That is on purpose. Instead of averaging every move into one score, it looks for the pattern type that keeps costing you the most.

The free check runs Stockfish. Stockfish is a free chess engine. It checks your last 10 games. It names your single most common recurring mistake in about 60 seconds. No account. No card. You connect your Chess.com or Lichess account, or upload a PGN file, which is just the standard file format chess games get saved in. It does not hand you a number to feel good or bad about. It hands you a name: the specific thing you keep doing, like walking into a fork, leaving a piece hanging, or letting your king's shelter fall apart.

Want the full picture? The paid report costs $10, one time. It covers 100 games instead of 10. It ranks every recurring mistake pattern you have by severity and by how often it happens. It comes with a fix course built around the top ones.

Think back to the two players from a moment ago. Both landed on the same average score for very different reasons. A single average cannot tell them apart. Naming the actual pattern behind the number can. That is the gap between watching a score and knowing what to work on next.

Run your last 10 games through the free check at https://blackboxchess.app/ and see which pattern is actually yours.