Guide
How your moves are graded
Every chess site will tell you a move was a blunder. Far fewer will tell you what that word means to them. These are the exact rules this analyser uses, printed from the same constants the engine reads — so the page cannot drift from the code.
Win percentage, not centipawns
Engines score a position in centipawns — hundredths of a pawn. It is a fine way to compare two positions and a poor way to judge a move, because the same number means different things in different places. Dropping a pawn from a level position can decide the game. Dropping one when you are already ten pawns up changes nothing you care about.
So every move here is measured in win probability: how likely the position was to be won before the move, and how likely after. The difference — the win percentage you threw away — is what the grade is based on.
This has a consequence worth stating plainly, because it surprises people: once you are completely lost, you stop blundering. There is nothing left to throw away, so further errors grade as inaccuracies. That is the honest answer, not a bug — a game already decided cannot be lost twice.
The eleven grades
The bands below are in win percentage points lost against the best move available. Four grades describe the move rather than score it: two of them reward finding the only move there was, and two — book and forced — mean no real decision was made, so they are left out of every average and rate.
| Grade | Win % lost | What it means |
|---|---|---|
| Brilliant | — | The only good move, and it gives up material to find it. |
| Great Move | — | The only move that holds the position together. |
| Best Move | — | The engine's own first choice. |
| Excellent | 0–5% | Not the top move, but it costs next to nothing. |
| Good | 5–10% | A reasonable move that keeps the position intact. |
| Book Move | — | Still inside known opening theory, so it was recalled, not found. |
| Inaccuracy | 10–20% | Slightly loosens the position. |
| Mistake | 20–30% | Hands the opponent a real advantage. |
| Miss | — | A won position let slip — still standing, but the win has gone. |
| Blunder | — | Past every band above: more than the last threshold, thrown away. |
| Forced | — | The only legal move. No decision was made, so none is graded. |
A move can qualify for more than one label; the harsher verdict wins. A sacrifice is only ever brilliant if you were not already losing when you played it, which stops a desperate throw in a lost game from being dressed up as inspiration.
How accuracy is worked out
Each move's accuracy comes from the win probability it gave away, on a curve where giving away nothing scores 100 and a catastrophe approaches zero. Turning those into one figure for the game is the part that is easy to get wrong:
- A plain average lets a long, quiet game bury a single decisive blunder among forty harmless moves.
- A harmonic mean does the opposite, and punishes one slip as though the whole game had been a disaster.
- So both are used: each move is weighted by how volatile the position was around it, and that weighted mean is averaged with the harmonic one. Sharp moments count for more, and one mistake costs what it actually cost.
The same method is used for both players in a game, so the two figures are always comparable with each other.
What this cannot tell you
It grades what a move cost, not why you played it. It does not know that you were in time trouble, that you had seen the idea and miscounted, or that you were playing for a trap. Some of that it can infer — the Insights tab reads the clock out of the PGN and can tell you whether your errors are the moves you played instantly — but the engine has no view on your reasoning.
It is also a search to a fixed depth, not an oracle. A deeper search sometimes disagrees, which is why the panel under the board keeps looking at the position in front of you and shows what it finds separately from the grade.
Common questions
- What is a blunder in chess?
- A blunder is a move that throws away a large part of your winning chances — here, more than 30 percentage points of win probability compared with the best move available. It is measured against what the position was worth before you moved, not against material, so giving up a queen for a forced mate is not a blunder.
- What is the difference between a mistake and an inaccuracy?
- Both cost you something; the difference is how much. An inaccuracy gives away 10 to 20 percentage points of win probability, a mistake gives away 20 to 30, and anything beyond that is a blunder. The bands are the same for both players and do not change with the time control.
- How is chess accuracy calculated?
- Each move is scored on how much win probability it gave away, and those per-move scores are combined into one figure for the game. Sharp positions count for more than quiet ones, so a long, calm game cannot bury a single decisive error, and one slip in an otherwise clean game is not punished as though it were ten.
- Why measure a move in win percentage instead of centipawns?
- Because a centipawn is worth a different amount depending on the position. Losing 100 centipawns from a level game can decide it; losing 100 from a position that is already winning by ten changes nothing. Win probability accounts for that, which is why a player who is already lost cannot keep registering fresh blunders here.
- Is this chess analysis free?
- Yes, entirely, and there is no account. The engine runs on your own processor inside the browser, so analysing a hundred games costs no more to provide than analysing one. Your games are never uploaded.
Try it on one of your own
Paste a PGN or pull a game from chess.com and every move is graded by exactly these rules. Open the board, or read why it all runs in your browser.