The evaluation bar is the most-watched number in chess and the most misread. Here is what the engine is telling you, and what it is not.
Turn on any analysis board and a number appears: +0.35, -2.10, M5. Players learn quickly that positive is good for White and negative is good for Black, and stop there. The number carries more information than that, and knowing how it is built tells you when to trust it.
An evaluation of +1.00 means the engine thinks White's position is worth about one pawn. That does not mean White is a pawn up. It means the whole position, material plus king safety plus piece activity plus structure plus everything else the engine weighs, comes out as roughly equivalent to being a pawn ahead with nothing else going on.
So a position where White is a rook down but mating can read +6.00, and a position where White is a pawn up with a shattered structure can read 0.00. The engine is pricing the position, not counting the pieces. Centipawns are just hundredths of that unit, which is why you see +1.20 rather than +1.2 pawns.
An evaluation says nothing about how hard the position is for a human to play. +1.50 in a simple endgame is close to won for a club player. +1.50 in a position where the only good plan is a five-move king walk is a coin flip in a blitz game. The engine assumes both sides play perfectly from here, which is exactly the assumption you cannot make about yourself or your opponent.
This is why a game can swing wildly between evaluations while both players feel they are doing something reasonable. The bar is measuring the position, not the difficulty.
Because a pawn means different things at different stages, most tools convert the evaluation into a percentage chance of winning. The conversion is deliberately steep near zero and flat at the edges. Between 0.00 and +1.00 the winning chances move a lot. Between +6.00 and +9.00 they barely move at all, because both are winning.
That single fact explains a lot of things players find odd. It explains why a review calls one mistake a blunder and shrugs at a bigger evaluation drop later in the same game. It explains why accuracy scores survive terrible endgame play in a won position. When you see an evaluation, ask what it does to the chances, not how many pawns it moved.
When the engine finds a forced mate it stops reporting pawns and reports a mate distance instead: M5 means mate in five moves for whoever is winning. M1 is not better than M5 in any practical sense, and neither is better than +30.00. They are all just winning. What matters is whether the mate is forced, and whether you can see it over the board.
Watch out for mate scores appearing and disappearing as the search deepens. A mate that shows up at depth 25 and vanishes at depth 28 was never there; the engine simply had not looked far enough to see the defence.
Every evaluation comes with a depth, which is roughly how many plies ahead the engine searched along its main lines. Depth 12 is a quick glance. Depth 20 is a serious look. Depth 30 is close to settled for most practical positions. The same position can read +0.30 at depth 14 and -0.80 at depth 24, and the deeper number is the better one.
Two practical consequences follow. First, do not treat a shallow evaluation as fact, especially in sharp positions. Second, when two tools disagree about your game, check their depths before assuming one of them is broken. Most disagreements are depth, not a bug.
There is also a horizon effect worth knowing about. An engine searching to a fixed depth can be fooled by a problem that sits just past its horizon, and will happily push it there with a delaying move. Deeper searches and quiescence rules reduce this, but sharp tactical positions are where shallow evaluations are least reliable.
The evaluation belongs to the engine's main line, the sequence it thinks both sides will play. If you look at only the number you lose the reason for it. A position at +2.00 with a forced sequence of six accurate moves behind it is not the same practical proposition as +2.00 that comes from a stable, easy advantage.
This is also why showing several candidate lines is useful. If the top three moves all score within a tenth of a pawn, the position has many good continuations and any of them is fine. If the top move is a pawn better than the second, that move is the position, and missing it is the mistake.
Statmate's analysis board shows the evaluation bar beside the position, the graph across the whole game, and a live panel with the engine's top lines for whatever position you are looking at. The panel reports its depth as it climbs, and you can cap it if you would rather stop the search early. Moves in the engine's lines are clickable, so you can walk down a continuation and come back.
The evaluation used for the report is fixed at one depth so the whole game is judged consistently, while the live panel keeps thinking as long as you sit on a position. When the two disagree the board tells you, rather than leaving you to notice it yourself.
The evaluation prices the whole position in pawn units, assuming perfect play from both sides. Convert it mentally into winning chances, respect the depth it was produced at, treat mate scores as a different scale, and read the line behind the number before you trust it.