Guides / What Brilliant, Great, Mistake and Blunder mean
What Brilliant, Great, Mistake and Blunder mean
Every move in a GChess Review game review gets one label. This page gives the exact rules behind each one, the same rules the app runs, so a label is something you can check rather than take on trust.
First, how a move is measured
The engine's score for a position, in pawns, is turned into expected points: the share of a point the side to move would score on average from there, from 0 (lost) to 1 (won), with 0.5 as even. The curve is the one Lichess fitted to millions of games:
expected points = 1 / (1 + e−0.00368208 × centipawns)
A move's cost is how many expected points it gave away compared with the position before it. Using expected points instead of raw pawns means that dropping a pawn in an equal position counts for much more than dropping one when you are already a rook up, which matches how games are actually won and lost.
The ordinary labels, by cost
- Best: the engine's own first choice. A move that is the only legal move is also Best.
- Excellent: costs less than 0.02 expected points.
- Good: costs less than 0.05.
- Inaccuracy: costs less than 0.10.
- Mistake: costs less than 0.20.
- Blunder: costs 0.20 or more, a fifth of a game result thrown away in one move.
Miss
A Miss is a different kind of error from a blunder: it is letting a win go. A move is a Miss when you were clearly winning (0.80 expected points or more) and the move leaves you below 0.55, or when you had a forced mate and the move no longer leads to one. A slower mate is not a Miss; giving up the mate is.
Great
A Great move is the only good move. It must be the engine's first choice, and the second-best move must be at least 0.18 expected points worse. Two exceptions keep the label honest: the game must not already be decided (above 0.90 for you), and a simple recapture of a piece that was just taken does not count, however forced it is.
Great can also go to the exact reply that punishes your opponent's Mistake, Blunder or Miss, when the next-best move would have let them off (a gap of 0.10 or more).
The app only awards Great when both moves were searched at full depth. Measured across repeat runs, a shallow search could move that 0.18 gap by up to 0.07, enough to switch the label on and off at random, so a shallow reading gets no Great at all.
Brilliant
A Brilliant move is a good sacrifice. All of these must hold:
- it is the engine's first choice;
- it gives up material, or leaves it to be taken;
- you still stand at least even afterwards (0.5 expected points or more);
- the game was not already decided, and you were not winning anyway with the second-best move (both above 0.90).
That last rule is why a famous sacrifice can be graded only Best. In Morphy's Opera Game the queen sacrifice is a beautiful finish, but Morphy was already winning by then without it, so it does not clear the bar.
Book
A Book move is a known opening move, played before the game left opening theory. It gets no grade, because it shows memory rather than calculation.
Accuracy
Each move also gets an accuracy from 0 to 100, from the change in win percentage, using the formula Lichess published: 103.1668 × e−0.04354 × (before − after) − 3.1669, plus one point of allowance for engine uncertainty, kept between 0 and 100. A move that loses nothing scores 100. A game's accuracy is the average of two means: one that gives more weight to sharp, important moves, and a harmonic mean that is pulled down by your worst moves. One terrible move therefore shows in the number, but it does not wipe out an otherwise good game.