What a centipawn is
Engines measure an advantage in pawns: +1.0 means you are roughly a pawn ahead, −2.0 means you are two pawns down. A centipawn is one hundredth of a pawn, so +1.0 is 100 centipawns. Using centipawns lets the engine express small differences precisely.
How centipawn loss works
For each of your moves, the engine compares the evaluation of the best move with the evaluation of the move you actually played. The difference is your centipawn loss for that move. Play the best move and your loss is 0; hang a knight and your loss might be 300+. It is always measured from your side, so a bigger number is worse.
Average centipawn loss (ACPL)
ACPL is simply the average of that loss across all your moves in a game. A low ACPL (say under 20) means a very clean game; a high ACPL means frequent or costly errors. Because it averages everything, one huge blunder and many perfect moves can still produce a moderate ACPL — which is why reviews also label individual moves as inaccuracy, mistake or blunder.
ACPL vs accuracy
Accuracy is a friendlier, capped 0–100% presentation of the same idea: low centipawn loss maps to high accuracy. ACPL is the raw, unbounded number some players prefer. To see yours, run a free game analysis, or set up a position on the analysis board to watch the evaluation — in pawns and centipawns — change move by move. For target ranges, see what is a good accuracy score.