The idea
What move classifications are
A move classification is a one-word verdict on a single move, produced by comparing what you played against what a strong engine, here Stockfish, considers best. The engine evaluates the position before and after your move, converts each evaluation into a win percentage, and grades the move by how much of that winning chance you kept or lost.
Most labels are pure loss thresholds: lose a little and it is good, lose a lot and it is a blunder. Two labels, brilliant and great, are earned rather than lost: they reward finding a specific kind of strong move. The eight classes below are exactly what a chessdesk review assigns, with the same symbols and colors you see in the product.
The eight classes
Every classification, from brilliant to blunder
The thresholds above are read straight from the review engine, so this page always matches what the tool assigns.
The scale
Why win percentage, not centipawns
Engines report a position in centipawns, hundredths of a pawn, where +100 means you are up about one pawn. The problem is that centipawns do not flatten out. Going from +900 to +1200 looks like the same three-pawn change as going from 0 to +300, yet the first swing changes nothing about a game you were already winning, while the second decides one that was balanced.
Win percentage fixes this by mapping the evaluation through a logistic curve onto your real chance of winning, from 0 to 100. Near equality small evaluation changes move the win chance a lot; once a position is winning the curve flattens and further gains barely register. Grading a move by lost win percentage therefore matches how much it actually cost you, which is why every classification here is measured on that scale.
The specific curve is the published Lichess win-percentage model. Chessdesk implements that fixed curve. Chess.com now uses a separate Expected Points model that also accounts for the player's rating.
Accuracy
How accuracy is computed
Accuracy turns the same win-percentage losses into a score out of 100. Each move gets a per-move accuracy: play the best move and you keep 100, and the score drops along a fitted curve as the win% you gave up grows. A small slip barely dents it; a blunder pulls it down sharply.
The single game accuracy is not a plain average of those per-move scores, because a plain average would let dozens of easy moves hide one game-losing error. Instead it blends two numbers:
- A volatility-weighted mean, which gives more weight to the sharp, swingy moments of the game where accuracy actually mattered and less to quiet moves where any reasonable choice was fine.
- A harmonic mean, which by its nature is dragged down by the worst moves, so a single blunder still shows up in the final number instead of being diluted.
Averaging those two gives a game accuracy that rewards steady precision in the critical phases while refusing to forgive a decisive mistake. This is the game-accuracy method documented by Lichess and implemented by Chessdesk. Accuracy is not standardized across chess sites, so a Chessdesk score should not be expected to match a Chess.com score.
Comparison
How this compares to Chess.com and Lichess
The sites share familiar words such as inaccuracy, mistake and blunder, but the labels are not interchangeable. They use different models, rules and analysis settings:
- Chess.com uses a similar ladder but adds Excellent between good and best and a Misslabel for failing to punish an opponent's error. Chess.com now publishes its Expected Points loss bands, and the model combines engine evaluation with the player's rating. Its special labels can also vary by rating. Read the official Chess.com classification documentation. Its documented loss bands are 0.00 to 0.02 for Excellent, 0.02 to 0.05 for Good, 0.05 to 0.10 for Inaccuracy, 0.10 to 0.20 for Mistake and 0.20 to 1.00 for Blunder.
- Lichess computer analysis marks inaccuracy, mistake and blunder and reports accuracy, but it does not hand out a brilliant tag there. Its win% and accuracy formulas are published, which is the model Chessdesk follows. Read the official Lichess accuracy documentation.
- chessdesk uses fixed win% thresholds you can read on this page: inaccuracy from 5%, mistake from 10%, blunder from 20%, plus the explicit brilliant and great rules above.
Sources last reviewed August 22, 2026. The Chess.com article is dated February 9, 2026.
FAQ
Frequently asked questions
What is a brilliant move in chess?
A brilliant move (marked !!) is a sound sacrifice. You give up at least 2 points of material, or leave a piece en prise, and the position still stays non-losing, meaning your win chance is at least 45% after the move, while the move is essentially the engine's best. What makes it brilliant is the combination of a real sacrifice that stays sound. It does not have to be the only move that works.
What is the difference between a great move and a brilliant move?
A great move (!) is the single best move in a critical position: it beats the second-best line by at least 18 win% points, so finding it is what holds the game together. A brilliant move (!!) is instead about a sound sacrifice of material. A great move is graded on being the only move that holds; a brilliant move is graded on giving up material and staying sound.
What counts as a blunder, mistake or inaccuracy?
The three are graded by how much winning chance a move throws away. An inaccuracy (?!) loses at least 5% win chance, a mistake (?) loses at least 10%, and a blunder (??) loses at least 20%. Because the scale is winning chance rather than raw evaluation, the same size of error matters more in a close game than in one that is already decided.
Why does chessdesk use win percentage instead of centipawns?
Centipawns measure material and position on a scale that never flattens out, so going from +9 to +12 looks as large as going from 0 to +3, even though the first change means nothing and the second decides the game. Win percentage maps the evaluation onto your actual chance of winning, which reflects how much a move really costs. A blunder in a balanced position and the same evaluation swing in a hopeless one are scored differently, the way a human would judge them.
How is chess accuracy calculated?
Each move gets a per-move accuracy from how much win percentage it gave up: play the best move and you keep 100, and the score falls along a fitted curve as the loss grows. The single game accuracy blends two averages of those per-move scores: a volatility-weighted mean that emphasizes the sharp, decisive moments, and a harmonic mean that refuses to let one bad blunder be diluted by many easy moves. This is the approach Lichess publishes and Chessdesk implements. Chess.com uses its own accuracy system.
Are these move classifications the same as Chess.com and Lichess?
No. Chessdesk uses fixed win-percentage-loss thresholds. Chess.com's Expected Points model combines engine evaluation with the player's rating, and its published cutoffs include Excellent, Good, Inaccuracy, Mistake and Blunder. Its Great, Brilliant and Miss labels add rating-sensitive rules beyond those cutoffs. Lichess publishes the win-percentage and accuracy formulas that Chessdesk implements, but Lichess computer analysis does not use Chessdesk's full label set.
Can a sacrifice be a blunder instead of brilliant?
Yes. A sacrifice only earns the brilliant tag when the position stays non-losing after it, with a win chance of at least 45%. Give up material and drop below that, and the move is graded like any other by how much it cost: if it throws away at least 20% win chance it is a blunder, not a brilliancy. The sacrifice has to actually work.
Keep going
See these labels on your own play: run a free game review, brush up on how moves are written in chess notation, or explore a position on the analysis board.
See your own brilliant moves
Run a free game review on your Chess.com or Lichess games and Stockfish will classify every move, brilliant to blunder, with accuracy and the best line you missed. No account needed.
Run a free game review