ChessAnalyzer Open the analyzer

Chess accuracy explained: what your accuracy score means

How ChessAnalyzer calculates accuracy from Stockfish 18 evaluations, why it differs from average centipawn loss, and how to use it.

Accuracy is the single number most players look at after a game. It answers one question: how close to the engine's best play were your moves? This guide explains how ChessAnalyzer computes it, what makes it fair, and how to use it to track your progress.

From moves to a score

For each of your moves, the review measures how many winning chances the move gave away compared with the best move. A perfect move loses nothing and scores 100. The more a move loses, the lower its score, following the curve popularised by Lichess:

move accuracy = 103.1668 × e−0.04354 × loss − 3.1669

Losing 5% of winning chances still scores about 80; losing 30% scores about 25. Book moves and forced moves are left out, because there was nothing to choose.

Forgiving difficult positions

Some positions are simply harder. When several moves are close together and the position is sharp, ChessAnalyzer forgives a little of the loss — up to 15% — so a small slip in a complicated middlegame counts slightly less than the same slip in a quiet position.

From moves to a game

A plain average would let one catastrophic blunder disappear among thirty good moves. So the game accuracy combines two averages: one weighted by how volatile each moment was, and a harmonic mean, which pulls the score down sharply when there is a big error. The result rewards consistency, as real games do.

Accuracy versus centipawn loss

Average centipawn loss (ACPL) is the older measure: the average number of hundredths of a pawn lost per move. It is still shown in reviews, but it has a weakness: losing three pawns in a position that is already won counts as much as losing three pawns in a level one. Accuracy fixes this by working in winning chances.

How to use it

Your accuracy also feeds the estimated rating.

Ready to try it on your own games?