Learn After the game Lesson 2 of 5
Reading the engine's numbers
Scores in pawns, mate scores and win chance, with the same board at 53%, 62%, 94% and 98% to compare.
One number for the whole board
When an engine has finished looking, it sums the position up as a single score, counted in pawns.
- A plus means White is better, and a minus means Black is.
- 0.0 is dead level.
- +1.0 means White is better by about as much as an extra pawn is worth. −2.5 means Black is better by two and a half.
The score always assumes that both sides play their best moves from here on.
Even the starting position has one. Moving first is worth a little, so it begins at about +0.3.
It isn't a count of the pieces
The pawn is only the unit. The score measures how good the position is, and material is one part of that.
Take a knight off Black's side of the starting position. By the usual values a knight is worth three pawns, so you might expect +3. The engine says +7.7.
That's because a whole piece up, with nothing given for it, is much more than three small advantages added together: it wins. The score grows faster than the material does.
It works the other way too. You'll see positions where one side has more and the score is 0.0, or worse.
Mate scores
When the engine has found a forced checkmate, it stops counting in pawns and counts moves instead.
M3 means the side to move can force checkmate in three moves, whatever the other side does. With a minus in front, −M3, it's Black who is giving the mate.
A mate score is the one kind that isn't an opinion. The engine has seen all the way to the end.
Try it: which score fits?
Three positions, and three scores to choose from for each. Check the sign first: who is better? Then the size.
From pawns to win chance
Scores in pawns have a flaw: the same step doesn't always mean the same thing. Going from 0.0 to +2.0 turns a level game into one you should win. Going from +8.0 to +10.0 changes nothing: you were winning, and you still are.
So this site turns every score into a win chance, a number from 0 to 100. Fifty is level, and 100 is a game that's as good as won. It's the bar under these boards: White's share from the left, Black's from the right.
| Score | White's win chance |
|---|---|
| 0.0 | 50% |
| +0.5 | 55% |
| +1.0 | 59% |
| +2.0 | 68% |
| +3.0 | 75% |
| +5.0 | 86% |
| +8.0 | 95% |
| A forced mate | 100% |
Look at how the steps shrink. The first pawn is worth nine points, and the pawns between +5 and +8 are worth three each.
It isn't a promise. A win chance of 75% doesn't mean you will win three games out of four from there. It's a measure of how much the position is worth, on a scale where equal steps mean about the same thing, and that makes it the right scale for asking what a move gave away.
The same board, with a little less on it
Here is the starting position again, three ways, each with one thing missing from Black's side.
First, a pawn. White's win chance goes from 53% to 62%: a real advantage, and nowhere near a won game.
Next, a knight instead. Now it's 94%. Between a pawn and a piece is the difference between being better and winning.
Last, the queen. She's worth three knights, and the bar moves four points, to 98%.
That's the lesson of the win chance: won is won. Once a game is decided, more material hardly changes it, and the moves that matter are the ones played while the bar is still near the middle.
Try it: whose game is it?
Five positions. For each, say who is winning, or whether it's about level. Counting the pieces will get some of them wrong, which is why the engine doesn't stop there.
After each answer the bar shows what Stockfish made of it.
How far to trust it
- Small differences mean nothing. The number shifts a little every time the engine looks deeper. 54% and 57% are the same position. Don't lose sleep between them.
- Big differences are real. When one move keeps 60% and another leaves 25%, something happened on the board, and it's worth finding out what.
- It assumes perfect play. The engine's 90% is against an opponent who misses nothing. Against a person, a "lost" position is often still a fight, and a "won" one still has to be won.
That last point is why a report here doesn't stop at Stockfish. A second neural network, trained on millions of games between people, says which moves players at your rating really find, so a report can show what a position is worth against the opponents you actually meet.