What Counts as Good Accuracy?
Accuracy scores are derived from centipawn loss by a formula each site invents. Why 95% in one game means less than 82% in another, and what the number is worth.
· 3 min read
Accuracy looks like an objective measurement because it is a percentage. It is not one. It is a presentation layer over centipawn loss, and the function that turns one into the other is invented independently by each site that shows it.
That does not make it useless. It makes it a number that only means something relative to itself.
Where the percentage comes from
Every implementation starts with the same raw material: how much value each of your moves gave away. It then maps that onto a nought-to-a-hundred scale using a curve of its own choosing, usually shaped so that small losses barely dent the score and large ones fall away steeply.
Three choices inside that mapping change the output:
- The curve. How harshly a 50-centipawn error is punished relative to a 200-centipawn one.
- The depth. A deeper analysis has stronger opinions about what you should have played, which usually lowers your score.
- The cap. Whether one disaster is allowed to count fully or is limited.
None of these is standardised. Comparing your accuracy on one site to a friend's on another is comparing two different measurements that happen to share a name.
Rough expectations in classical chess
| Accuracy | Usually means |
|---|---|
| 95%+ | A short game, a forced sequence, or genuinely excellent play |
| 90 to 95% | Very strong, few real decisions gone wrong |
| 80 to 90% | A good game at club level |
| 70 to 80% | Normal for an improving player |
| under 70% | Several moves worth studying |
Now the caveat that undoes most of the table. A twelve-move game where your opponent blundered into a mate can score 98%. A ninety-move endgame grind where you found difficult moves for hours might score 84%. The second game was harder and better played.
Length and sharpness dominate
Two structural facts drive accuracy more than skill does.
Short games score higher. Fewer moves means fewer chances to lose value, and the opening moves you played are usually theory, which costs nothing.
Quiet games score higher. In a position with one sensible plan, most moves lose nothing. In a sharp position with four candidate moves and real differences between them, ordinary human choices cost centipawns.
This means accuracy partly measures what kind of game you had, not only how well you played it. A player who steers into simple positions will post better numbers than one who enters complications, at equal strength.
The one thing it is good for
Your own trend, on one site, at one time control. That comparison holds every variable constant except your play, and it is the only version of the number that answers a real question.
Anything else, including a single game read in isolation, is a statistic in search of a meaning. The individual moves that lost the most are the part of the report worth your time, and they are sitting underneath the percentage.
Why the same game scores differently
Take one game and analyse it three ways. The differences are not small.
| Setting changed | Effect on the score |
|---|---|
| Depth 12 to depth 20 | Usually drops 3 to 8 points |
| Single-move loss capped at 300 vs uncapped | Can differ by 10 points in a game with one disaster |
| A steeper conversion curve | Shifts every score, most at the low end |
None of those is a bug. They are three defensible choices producing three defensible numbers for identical play. It is the reason a screenshot of an accuracy score settles no argument about how well someone played.
Two scores worth distrusting
A very high score in a short game. Twelve moves of theory followed by an opponent's blunder is not a demonstration of accuracy, but it produces a spectacular one. Anything under about twenty-five moves should be read with that in mind.
A high score in a game you found difficult. If the position felt sharp and the score came back at 93%, one likely explanation is that the sharpness was in your head rather than on the board, and another is that the analysis was shallow. Deepen it and see which.
The score that means something is a mid-length game in a familiar time control, compared against your own history. Everything else is noise dressed as a measurement.
Common questions
- What is a good accuracy in chess?
- In a classical game, above 90% is very strong, 80 to 90% is a good game for a club player, and 70 to 80% is normal at intermediate level. The figure is much lower in fast time controls and much higher in short or forced games, so it means little without that context.
- Why is my accuracy different on different sites?
- Because the formula converting centipawn loss into a percentage is not standard. Each site chooses its own curve, its own analysis depth and its own cap on single-move losses. The same game genuinely scores differently, and neither number is incorrect.
Try it on your own position
The analysis board runs the same engine in your browser, free. The app adds what a web page cannot: reading a position straight from a photograph, saved boards, and analysis with no connection.
