Open any post-game report on Chess.com or Lichess and one number stares back at you: average centipawn loss. People screenshot it, brag about it, and quietly panic over it. Yet most adult improvers could not tell you what it actually measures, what a “good” value is for their rating, or why chasing a lower one sometimes makes them play worse. This is the explainer that fixes that.
What average centipawn loss actually measures
A centipawn is one hundredth of a pawn. It is the unit engines use to express an evaluation: +1.00 means “white is up the equivalent of one pawn,” and +0.35 means “white is up 35 centipawns.” When you play a move, the engine compares the evaluation before and after. If the position was +0.50 for you and your move dropped it to +0.20, you gave up 30 centipawns on that move. That single-move figure is your centipawn loss for the move.
Average centipawn loss (ACPL) is simply the mean of those per-move losses across the whole game. Lower is better, because it means your moves stayed closer to the engine’s top choice. That is the entire concept. Everything difficult about ACPL comes not from the definition but from the word “average.”
Why the average hides more than it shows
Averages flatten the story. A game where you played 39 near-perfect moves and one catastrophe that hung your queen can produce the exact same ACPL as a game where you drifted 15 centipawns off on every single move. Those are completely different problems. The first is a blunder-control problem; the second is a positional-understanding problem. ACPL treats them as identical, which is why it is a useful thermometer but a terrible diagnosis. It tells you that something is wrong, not what.
It is also engine-dependent and depth-dependent. Chess.com and Lichess use different engines, different search depths, and different rounding, so an ACPL of 25 on one platform is not directly comparable to 25 on the other. Treat the number as an internal trendline for yourself, not as a universal grade.
The ACPL number to chase at your rating
People desperately want a target. Here is an honest one, based on the ranges typically seen in analyzed rapid and classical games. These are approximate and will shift with time control, but they are close enough to calibrate expectations.
- Under 1000: ACPL of roughly 90–130. Games are decided by hanging pieces, not subtle inaccuracies.
- 1000–1400: roughly 60–90. Fewer outright blunders, but still one or two per game.
- 1400–1800: roughly 40–60. Blunders become occasional; the losses come from inaccurate plans.
- 1800–2200: roughly 25–40. Clean games with the odd slip under pressure.
- 2200+ and titled players: often below 20 in classical, sometimes single digits in a clean win.
The single most useful takeaway: at your level, the fastest way to cut ACPL is not to find deeper positional ideas. It is to stop giving away 300-centipawn chunks in one move. One prevented blunder is worth more to your average than ten slightly-improved quiet moves.
Why a low ACPL can be a trap
Here is the counterintuitive part almost no one tells you. You can lower your ACPL by playing more passively, and that will not make you a stronger player. Quiet, symmetrical positions with few forcing lines are easy to navigate at low error, so a player who avoids complications, trades early, and shuffles pieces can post a flattering number while never practicing the skills that actually raise a rating: calculation, initiative, and converting winning positions.
Sharp, double-edged games naturally produce higher centipawn losses because the punishment for a slightly-off move is larger and the tree of reasonable moves is wider. A fighting draw against a stronger opponent might show a higher ACPL than a lifeless win, yet teach you ten times more. Do not optimize the metric at the expense of the skill the metric is supposed to proxy. ACPL is a rear-view mirror, not a steering wheel.
The three things inflating your ACPL, and how to fix each
1. The single catastrophic move
For most players under 1600, one move accounts for the majority of a bad game’s total loss. Hang a rook and you have banked 500 centipawns in a single ply, which an entire game of clean play cannot offset. The fix is not “play better generally.” It is a specific pre-move checklist: before every move, ask what your opponent’s last move attacks and whether your intended move leaves anything undefended. This one habit does more for ACPL than any opening study.
2. The lost-position spiral
Once you are losing, centipawn loss inflates fast because every reasonable defensive try still loses ground against best play, and the engine keeps punishing you for a position that is already bad. A single early mistake can drag the rest of the game’s numbers down with it. The lesson is that ACPL rewards not being lost in the first place. It also means you should read your game report by finding the first big swing, not by obsessing over the pile of red moves that followed it.
3. Time trouble
The last five moves before a time scramble are where accuracy collapses. If your ACPL is fine through move 30 and then explodes, you do not have a chess problem, you have a clock-management problem. Track when your errors cluster. If they bunch at the end, the fix lives in your time budgeting, not your calculation.
A practical routine to bring ACPL down
Chasing the number directly is the wrong approach. Chase the behaviors that produce it, and let the number follow.
- Blunder-check every move. A two-second scan for undefended pieces and opponent threats. This alone can cut a beginner’s ACPL by a third.
- Review the first mistake, not the last. In each analyzed game, find the earliest move where the evaluation swung by 100+ centipawns and understand only that one. Fixing the source beats mopping up the consequences.
- Separate blunder-games from drift-games. If a bad ACPL came from one disaster, train tactics and board vision. If it came from many small losses, train pawn structures and planning.
- Ignore ACPL in sharp practice games. When you are deliberately playing aggressive, complicated positions to build calculation, expect a higher number and do not treat it as failure.
If you want to go deeper on turning a game report into an actual training plan rather than a guilt trip, our blame-free post-mortem system walks through exactly how to read the swings without drowning in engine lines.
ACPL versus accuracy score: same data, two different stories
Chess.com’s “Accuracy” percentage and ACPL are built from the same underlying centipawn losses, just presented differently. Accuracy runs a formula that converts your losses into a 0–100 score, which is friendlier to read but compresses the extremes: the gap between a 92 and a 95 can hide a meaningful difference in play, while both feel “good.” ACPL is the rawer, more honest cousin. If you want to track improvement over months, ACPL trends are usually more informative than accuracy percentages, precisely because they are less smoothed.
Either way, remember that these scores depend heavily on the tool computing them. Cloud engines and paid analysis platforms run deeper searches and will report different numbers than a quick browser analysis. We compared what that extra depth is actually worth for improvers in our Chessify cloud-engine review, and we pressure-tested how much of a platform’s per-move feedback is signal versus noise in our Aimchess review. The short version: more decimal places do not automatically mean more insight.
The bottom line
Average centipawn loss is a genuinely useful number when you treat it as a trendline for your own play and a flag that something went wrong. It becomes misleading the moment you treat it as a grade to be maximized, compare it across platforms, or let it push you toward safe, passive chess. Watch the trend, hunt the first big swing in every game, kill your blunders before you polish your quiet moves, and the number will take care of itself.
Leave a Reply