--:--
--:--
Game Report
Free Game Report
Every move of one Chess.com or Lichess game classified by Stockfish, with an evaluation graph and accuracy for both sides. The engine runs in your browser.
--:--
--:--
Every move of one Chess.com or Lichess game classified by Stockfish, with an evaluation graph and accuracy for both sides. The engine runs in your browser.
Game Report takes one Chess.com or Lichess game and analyses every move with Stockfish, producing a classification for each move, an evaluation graph across the game, and notes on the moments that changed the result. The engine runs in your browser as WebAssembly, so the game is analysed on your own machine rather than queued on a server.
The graph plots the engine's evaluation after every move, from White's point of view. A flat line means neither side is making progress; a sharp step means something concrete changed. Those steps are what you are looking for — a game is usually decided by two or three of them rather than by gradual drift.
Evaluations are given in centipawns, where 100 is roughly the value of one pawn, or as a mate score when a forced mate exists. A reading of +1.5 means White stands about a pawn and a half better with best play from both sides. It is a statement about the position, not a prediction: engines evaluate assuming both players find the best continuation, which at most levels is not what happens next.
Each move is compared against what the engine considers best in that position, and the size of the gap decides the label. A blunder is a move that changes the assessment of the game — winning to equal, or equal to losing. A mistake costs real ground without necessarily deciding anything. An inaccuracy is a small step in the wrong direction, often a reasonable-looking move that misses something better.
Two things are worth keeping in mind when reading them. First, a move flagged as an error is not necessarily a move you should have found — engine best play routinely involves lines no human would calculate over the board. Second, the label describes the outcome, not the thinking. A blunder played after careful calculation and a blunder played instantly look identical in the report but mean very different things about what you need to work on.
Accuracy converts evaluation loss into a percentage. Rather than counting how many moves were engine-best, it weights each move by how much the position actually changed, using a win-probability model: a move that drops the evaluation from +9 to +6 barely affects your chance of winning and is scored gently, while one that moves the position from equal to lost is penalised heavily.
This is why accuracy tracks results better than a raw best-move count. It also means accuracy is not comparable across games — a sharp tactical game and a quiet positional one produce different numbers for the same quality of play.
Start with the largest swing in the graph rather than reading move by move. Find the position before your worst move and work out what you would need to have noticed to avoid it. That is more useful than knowing the engine's preferred continuation twelve moves deep.
One game will not tell you what your weaknesses are — every player has bad games, and a single report cannot separate a pattern from an accident. To see what repeats, run a Profile Report, which analyses a batch of recent games and reports where your losses actually come from. If the same tactical motif keeps appearing, train it directly with Lumichess puzzles, or work through your own missed positions with mistake training.