ChessGrader / Guides / Move classifications

Chess move classifications, explained

Every graded chess move gets one of 11 labels, from Brilliant down to Blunder. Here's what each label means, the rating-aware expected-points boundaries behind ordinary moves, and why the same move can get different labels on different sites.

By the ChessGrader team · Updated July 10, 2026

ChessGrader follows the public shape of Chess.com's Classification V2 model. The quick version: Excellent loses at most 2 expected points, Good stays below 5, an Inaccuracy starts at 5, a Mistake at 10, and a Blunder at 20. Player rating affects how engine evaluation maps to expected score, and centipawn and mate safeguards handle the places a probability curve cannot.

All 11 labels at a glance

Each grade in a bit more detail

Brilliant (!!)

The engine's best move that is also a genuine sacrifice — material given up by the raw capture arithmetic, in a position you hadn't already won, and still sound afterwards. It's the rarest label by design; many games have none. Full breakdown on the brilliant moves page.

Great (!)

A move that had to be found. Either it was the only good move — the second-best option at least 18% worse in win probability — or it was the precise move that punished an opponent error which had turned a lost or equal game. Like Brilliant, it can only be awarded after deep verification.

Forced (□)

The position offered exactly one legal move. Forced is descriptive rather than praise: finding the only legal reply is different from choosing the best move among real alternatives.

Best (★)

You played the engine's top choice, a move that ties its second line within 10 centipawns in a non-mating position, or a line with the same mate distance. At that margin the engine itself can't reliably tell the non-mating lines apart, so you get full credit. Strong players collect a lot of these.

Excellent

Not the engine's first choice, but it gave up no more than 2 expected points. In practice the difference between Best and Excellent rarely matters over the board.

Good

A reasonable move that gave up more than 2 but fewer than 5 expected points. Your position is a bit worse than it needed to be, and that's all.

Book

A low-loss move in the game's still-contiguous opening prefix, up to move 8. Once a move leaves that prefix, the Book label does not restart later in the game.

Inaccuracy (?!)

You lost at least 5 expected points. One inaccuracy is background noise; a string of them is how equal positions quietly become lost ones.

Miss (✗)

The label most sites don't have. It covers skipping mate on the spot, losing another forced mate while still winning, missing a concrete tactic found in the best line, or immediately giving back a large opportunity your opponent just handed you. A Miss can look modest numerically; what matters is the specific chance that was sitting there.

Mistake (?)

At least 10 expected points lost. A mistake changes the character of the game — an edge becomes equality, equality becomes trouble.

Blunder (??)

At least 20 expected points lost. This is the game-changer: a winning position becomes equal, or an equal one becomes lost. In already-decided positions, ChessGrader caps the centipawn backstop below Blunder rather than pretending the result was lost twice.

How forced mates are handled

Mates break centipawn math — "mate in 3" isn't a number you can subtract from. Most tools handle this crudely. ChessGrader uses three explicit rules:

• Skipping mate in one is a Miss. That stays true even when the move keeps another forced mate: mate on the board was the concrete chance.
• Other longer routes keep an Excellent floor. If mate remains forced and there was no mate on the spot, choosing a slower route is not treated as a numerical catastrophe.
• Losing a forced mate while still winning is a Miss, not a Blunder. You missed the kill; you did not necessarily throw the game.
• A defender already facing forced mate isn't charged extra. When every move loses, no move is a blunder. You don't get punished for the inevitable.

Why thresholds use expected points plus a centipawn backstop

A 100-centipawn loss in an equal position turns a draw into a loss; the same 100 centipawns at +9.00 changes nothing. Raw centipawn thresholds would call both the same. ChessGrader converts evaluation and player rating into an expected score, then measures the expected points lost. A centipawn backstop still catches bad technique after that expected-score curve has flattened, with a cap that prevents an already-decided position from manufacturing a fresh Blunder. The complete logic is on the methodology page.

Are labels the same everywhere?

No. Chess.com, Lichess, and ChessGrader all grade with Stockfish, but the label sets and thresholds differ. Lichess only marks Inaccuracy, Mistake, and Blunder. Chess.com uses a fuller vocabulary and a rating-aware Expected Points Model. ChessGrader follows the public V2 boundaries, but Chess.com's fitted model and search settings remain private, so no fixed parity guarantee is honest. ChessGrader's approximation and safeguards are public, so disagreements can at least be inspected. To see them applied to your own games, run a free game review.