A chess evaluation expresses the engine's assessment of a position, in pawns, from White's point of view. Positive numbers favour White and negative numbers favour Black, so +1.20 means White is about a pawn and a fifth ahead. A score written M3 means White forces mate in three moves, and -M3 means Black does.
The number beside the board is the engine's opinion about who is winning and by how much. It is the single most useful piece of information in computer chess, and also the most commonly misread, because its precision is not the same as certainty.
What the number measures
Evaluations are given in pawns, from White's perspective. Positive means White is better, negative means Black is better. Zero means the engine sees no advantage either way.
| Score | Roughly means |
|---|---|
| 0.00 | Dead level, or a position nobody can make progress in |
| +0.30 | A small pull for White, often just a slightly better structure |
| +1.00 | About a pawn up, or a comparable positional advantage |
| +3.00 | A piece for a pawn, or a clear winning attack |
| +8.00 | Completely winning by material |
| M5 | White forces checkmate in five moves |
Those bands are rules of thumb. In an endgame, half a pawn of advantage can be completely winning because there are few pieces left to defend with. In a sharp middlegame, plus one can be a temporary claim that evaporates in two moves. The number is a guide to the position, not a verdict on it.
Mate scores are different
A forced mate is written with an M: M3 means mate in three, and -M3 means the opponent mates in three. Mate scores are not large centipawn numbers, because the distance to mate matters. Mate in two is better than mate in nine, and a single number cannot express that.
Why the number moves as the engine thinks
A search starts shallow and deepens. At low depth an engine may think a sacrifice wins, then find the defensive resource a few plies later and revise the score downwards. Watching the evaluation settle is normal. Watching it swing wildly usually means the position is genuinely sharp or the search is still very shallow.
The number is relative to the position, not the player
An evaluation says nothing about whether a human in a given time control can hold the position. Once a narrow position is winning, the practical question becomes how hard it is to play, and a plus one position with one line and no margin is harder than a plus one position with several ways to go wrong.
How to use an evaluation well
- Look at the direction of travel across a game, not just the current number. Where the bar crosses from one side to the other is where the game turned.
- Compare your move's resulting score with the best move's score. The gap is what your move cost, and that is more informative than the score itself.
- Read the line that produced the score. A number without its principal variation is a claim with the evidence removed.
- Treat any score as specific to the depth and settings that produced it. Annotate a game at low depth and you will grade moves that a deeper search would call fine.