Deblunder study notebook
Chess accuracy score: what it means
Accuracy is a useful summary, not a grade for your chess. It estimates how closely your moves followed the engine's preferred play in one game. Two platforms can score the same game differently, and a high number can hide the one decision that lost the game. This guide explains what the score measures, what it cannot tell you, and how to use it as one signal among several.
The short definition
An accuracy score compares the moves you played with the moves an engine preferred in the same position. A move that matches the engine's top choice is treated as more accurate than a move that loses material, allows a tactic or damages the position. The final number is usually scaled from 0 to 100, where higher means closer to the engine's choices.
That definition already contains the limits. Accuracy depends on the engine, its depth, the position and the scoring formula. It is not a rating, it is not a measure of chess understanding and it is not comparable across platforms without care. It describes one game through one engine's eyes.
How Chess.com calculates accuracy
Chess.com's current score is powered by CAPS2, an updated version of its Computer Accuracy Precision Score. Chess.com says moves are still compared against top engine recommendations, but the math changed to spread scores in a way that feels more like a school grade. The company reports that most scores now fall between 50 and 95, illustrating the scale with players rated 1000 to 1500.
The exact formula is not public. That matters when you read advice claiming a precise threshold such as 80 being good at every level. Chess.com's own support article treats accuracy as a measurement of how closely you played to the engine's best play, not as a rating predictor. The article also notes that CAPS2 replaced the older CAPS scale, so older blog posts that quote CAPS thresholds may no longer describe the number you see.
How Lichess calculates accuracy
Lichess runs computer analysis with Stockfish and reports average centipawn loss, usually shortened to ACPL. A centipawn is one hundredth of a pawn, so losing 100 centipawns is roughly losing one pawn of evaluation. Lichess explains in its FAQ that the top computer move loses zero centipawns and lesser moves lose more, and that fewer centipawns per move indicates stronger play.
Lichess also shows an accuracy figure per game. Its precise aggregation is a Lichess implementation detail, so do not assume it uses the same scale or thresholds as Chess.com. If you track accuracy over time, keep a separate baseline for each platform.
Why the same game can get different numbers
The same PGN can produce different accuracy scores on different sites. Engines may search to different depths, use different evaluation scales, apply different opening books or classify moves with different thresholds. A move that one tool calls a mistake can fall inside another tool's acceptable range. Even on one platform, a settings change or a new formula version can shift the number.
The practical consequence is simple: compare your accuracy with your own recent accuracy on the same platform and the same time control. Do not compare your number with a friend's number from another site, and do not compare a 2023 screenshot with a 2026 score.
- Different engine depth and settings.
- Different move-classification thresholds for inaccuracy, mistake and blunder.
- Different treatment of opening theory and forced moves.
- Different time controls and opponents, even within the same platform.
- A small sample, where one sharp position dominates the average.
What is a good accuracy score?
There is no universal number that means good chess. A 70 in a complex, sharp game against a stronger opponent can represent better decisions than an 85 in a quiet game where most moves were obvious. Chess.com's own 50 to 95 band illustrates how compressed the scale is for ordinary players; it is not a target you must beat.
A more useful question is whether your accuracy is stable and whether the errors inside the game are the same ones that cost you points. A player who scores 75 with one catastrophic blunder has a different problem from a player who scores 75 through many small inaccuracies. The average hides that difference.
What accuracy hides
Accuracy is an average over moves, and chess is not decided by an average. One move that turns a winning position into a loss can matter more than twenty harmless moves. The score also says little about whether you found the right plan, managed your clock or converted a better endgame.
It can also reward passive play. If you avoid complicated positions, you may make fewer clear errors while never creating serious problems for the opponent. A high accuracy score in a drawn, passive game is not evidence that your chess improved.
- The criticality of the position: not all moves carry the same weight.
- The number of blunders, independent of the average.
- Conversion of winning positions and defence of worse ones.
- Time trouble and decision speed.
- The sample size behind a trend.
How to use accuracy without fooling yourself
Track accuracy in a fixed context: same platform, same time control, at least twenty games, and compare the median rather than a single best or worst game. Then open the games behind the number and group the causes. Did you lose material to the same hanging piece? Did you miss a forcing reply? Did you spend too little time on critical positions?
Use accuracy as a pointer to the games worth reviewing, not as the review itself. A drop in your median may tell you that something changed. It does not tell you what changed. The explanation comes from the board, from your notes, and from comparing positions where the same relation appears.
This is why game analysis starts with your own explanation before the engine. The number can tell you where to look; the cause tells you what to train.
A better set of signals
If accuracy is your only metric, you are optimizing a summary. Combine it with signals that describe the decisions behind it. The exact numbers depend on your sample, so treat them as counts and comparisons, not universal benchmarks.
- Blunders per game, counted separately from inaccuracies.
- The recurrence of one named cause across a sample of games.
- First-attempt performance on positions taken from your own games.
- Results in positions where you were clearly better.
- Time spent on critical decisions, when clock data is available.
How Deblunder uses accuracy
Deblunder does not reduce a game to one accuracy score. It uses Stockfish to find where the evaluation changed, then looks for the board relation that explains the mistake: a piece that was defending or blocking, a check that was missed, a line that opened. When the position proves the cause, it shows the evidence. When it does not, the cause stays unknown instead of receiving an invented explanation.
Training measures your first valid attempt on positions from your own games. Later practice does not rewrite that result. Over a sample, you can see whether the cause you trained appears less often, while the sample size remains visible. That is a narrower and more actionable signal than a single accuracy number, and it is not a promise of rating gain.
If you want the full method, read how Deblunder works. If you are choosing between platforms, chess analysis tools compared explains what each one does well.
Frequently asked questions
What does an accuracy score measure?
It estimates how closely your moves followed the engine's preferred moves in one game. It is a summary of engine agreement, not a rating or a complete measure of chess skill.
What is a good accuracy in chess?
There is no universal threshold. It depends on the platform's scale, the engine, the time control, the opponent and the complexity of the position. Compare your own median over a stable sample instead of chasing a number.
Why do Chess.com and Lichess give different accuracy scores?
They use different formulas, engine settings and move-classification thresholds. Chess.com uses CAPS2, and Lichess reports its own accuracy and average centipawn loss. The same moves can therefore produce different numbers.
Can I compare accuracy between time controls?
Only with caution. Rapid and bullet create different decision conditions, and one sharp position can dominate a short game. Keep separate baselines and compare like with like.
Does a high accuracy score mean I played well?
Not necessarily. Accuracy is an average and can hide one decisive blunder, passive play or a failure to convert an advantage. Review the critical moments behind the number.