How every move is classified
Most review tools show you a label and leave you to guess what it meant. These are the actual numbers this one uses.
Everything starts with winning chances
A raw engine evaluation is a poor measure of how bad a move was. Going from +9 to +7 is barely a scratch; going from +0.3 to −1.7 can lose the game. So the evaluation before and after each move is first converted into a winning percentage — how often the side to move would expect to win from that position — using the same logistic curve Lichess publishes.
The difference between those two percentages is the move's loss: how much winning chance it gave away. Every label below is a band of that one number.
The labels
| Label | Winning chance lost |
|---|---|
| Best | the engine's own move, losing 1 point or less |
| Only move | the engine's move, where the second-best was 12 or more points worse |
| Excellent | 2 points or less |
| Good | 5 points or less |
| Inaccuracy | 10 points or less |
| Mistake | 20 points or less |
| Blunder | more than 20 points |
Only move is worth explaining. It is not a better version of "best" — it marks a position where there was exactly one road and you found it. If the second-best move would have cost you 12 or more percentage points, the position was on a knife edge, and playing the right move there deserves noting.
What makes a move brilliant
"Brilliant" is the label most likely to be handed out too freely, so it has four conditions and all of them must hold:
- Real material is given up. At least 1.5 pawns' worth, measured after the opponent's best recapture — so an ordinary trade where the material comes straight back does not qualify.
- The move is sound. It must cost 2 percentage points or less against the engine's choice. A sacrifice that does not work is not brilliant.
- You are still doing fine afterwards. At least a 45% winning chance once the dust settles. Below that the sacrifice is just losing.
- You were not already winning easily. If you were more than 3 pawns up, giving material back is slop, not brilliance.
That last condition is deliberately measured in material rather than winning chances. A forced mate already scores 100% before it is played, so a percentage-based guard would throw out every sacrifice that leads to mate — which is most of the good ones.
How accuracy is calculated
Each move's loss is converted into a per-move accuracy score on a curve, and your accuracy for the game is the average of those scores across your own moves. The curve is steep at the start and flattens out: the difference between losing 0 and 5 percentage points matters far more than the difference between losing 40 and 45, because by then the damage is done.
This is the same curve Lichess uses, so accuracy figures here are comparable with the ones you see there.
Performance rating
Alongside accuracy, each side gets a rating that this game's play was worth — roughly, the strength a player would need to have to play this well over a game of this length. It is a single game, so treat it as an indication rather than a verdict. Ten games tell you far more than one.