After a game, most chess apps color your moves: green for good ones, yellow for inaccuracies, orange for mistakes, red for blunders. The labels feel objective, like a grade on a test. They are useful — but only if you understand what is being measured. Otherwise you end up worrying about moves that didn't matter and ignoring the ones that decided the game.
This guide explains how those labels are produced, why the thresholds are based on winning chances rather than raw pawn values, and how to read a review so it actually improves your play.
Step one: the engine evaluates every position
A game review starts by running an engine — on ChessDigi, Stockfish running in your browser — on every position of the game. For each position it produces an evaluation, usually shown in pawns: +1.5 means White is better by about a pawn and a half; −3.0 means Black is better by about three pawns; a number like "M4" means a forced mate in four.
It also finds the best move in each position. Your move is then compared with the best move by looking at how the evaluation changed.
Step two: convert the evaluation to winning chances
Here is the key idea that most players never hear about. A drop of one pawn does not mean the same thing in every position:
- At 0.0 (equal), losing a pawn's worth of evaluation turns an equal game into a clearly worse one. That matters a lot.
- At +8.0 (completely winning), dropping to +7.0 changes nothing. You are still winning easily.
So modern reviews don't grade moves by pawns lost. They first convert the evaluation into an expected winning chance — roughly, how often a player in that position would win — using a curve that flattens out at the extremes. Then they grade the move by how much winning chance it gave away.
On ChessDigi, the grades are based on how many percentage points of winning chance a move costs:
| Label | Winning chance lost |
|---|---|
| Best move | the engine's own top choice |
| Excellent | less than 2 points |
| Good | 2 to 5 points |
| Inaccuracy | 5 to 10 points |
| Mistake | 10 to 18 points |
| Blunder | 18 points or more |
Other sites use slightly different cut-offs and names, but the principle is the same everywhere.
An example from a real opening trap
The Blackburne Shilling Gambit is a dubious opening trick that has fooled club players for over a century. It gives a clean demonstration of all three labels.
Follow the evaluation:
- 3...Nd4 leaves White about a pawn better. Objectively dubious — an inaccuracy or worse — but it sets a trap.
- 4.Nxe5? looks natural (a free pawn and an attack on f7), but after 4...Qg5 Black attacks the knight and the g2 pawn at once. The evaluation swings by almost two pawns, from a comfortable White edge to a worse position. That is a mistake.
- 5.Nxf7??, going for the rook, loses outright: 5...Qxg2 and Black's attack crashes through, with the evaluation dropping to about −6.6. That is a blunder.
Notice that the "free pawn" on move 4 and the "free rook" on move 5 look similar to a beginner, but their grades are very different — because one gave away a modest advantage and the other gave away the whole game.
How accuracy percentages work
The accuracy number shown at the top of a review summarizes the whole game. Each move is scored based on the winning chance it lost — a best move scores near 100, a blunder near 0 — and the scores are averaged for each player.
Two things follow from this:
- Accuracy depends on the game, not only on you. A quiet, simple game produces high accuracy for both sides. A sharp, tactical game full of hard decisions produces lower accuracy even for strong players.
- Accuracy in a lopsided game is misleading. If your opponent blundered on move 10 and you converted slowly, you might have mediocre accuracy and an easy win. If you were lost from the opening, your accuracy may look fine simply because every move in a lost position loses little.
Use accuracy to spot trends over many games, not to judge a single game.
Which labels deserve your attention
A typical club game might show a dozen inaccuracies, three or four mistakes and one or two blunders. Trying to understand all of them is overwhelming and mostly useless. A better order:
- Blunders first. One or two blunders decide most games under 1800. Find the one that swung the result.
- Then mistakes that repeat across games. If the same kind of mistake — say, pushing kingside pawns too early — shows up again and again, it is a habit worth fixing.
- Inaccuracies last, and only in the opening. Opening inaccuracies can reveal gaps in your repertoire. Middlegame inaccuracies are usually small differences between reasonable plans.
When the engine's "best move" isn't the right lesson
Sometimes the engine's top move is a long, precise sequence no human would find. If your move was the natural, safe choice and it lost only a little, don't treat it as a failure. The better question is: was there a simple move that was clearly better? If the answer is no, move on.
It also helps to review before turning on the engine: note the two or three moments where you felt unsure, then see how the engine judged them. The moves you were unsure about teach far more than the ones you never thought twice about.
Try it on your own game
The fastest way to understand the labels is to see them on a game you just played. Play a game on ChessDigi against the computer at your level, and when it ends, tap See why you lost or Review game. The review lists your costliest moves and draws arrows for the better alternatives. For a full routine on reviewing games, read how to analyze your games.
Frequently asked questions
What is the difference between a mistake and a blunder?
The size of the damage. A blunder gives away a large part of your winning chances — often turning a win into a draw or a loss. A mistake is a smaller but still significant loss. The exact cut-off varies by site.
Why was my move marked as a blunder when I only lost a pawn?
Because in that position the pawn mattered a lot. In an equal position, losing a pawn can cut your winning chances substantially. The same pawn loss in a position where you are already winning by a lot would barely register.
Is 80% accuracy good?
It depends on the game. In a calm game against a similar opponent, 80% is solid. In a sharp tactical game, it can be very strong. Compare your accuracy across many games of the same type rather than looking at one number.
Should I trust engine labels completely?
Trust them to tell you where the evaluation changed, not always why. The engine sees tactics perfectly but cannot explain plans. Use its labels to find the critical moments, then think through those moments yourself.