Reading the evaluation bar

The vertical bar beside the board, the number above it, and the captured pieces below.

The evaluation is the single most useful thing on the screen and the easiest to misread. It is not a prediction of who will win, and it is not a score you accumulate. It is the engine's opinion about one position, expressed in a unit borrowed from material.

Centipawns

A centipawn is one hundredth of a pawn. The readout divides by a hundred and shows two decimals, always signed from White's point of view:

The sign convention matters: the number is always from White's perspective, whichever side you are playing and whichever way the board is facing. A move by Black that improves Black's position makes the number go down. This is why the bar is often easier to read than the figure — you do not have to remember which direction is good for you.

"A third of a pawn better" does not mean anyone is winning. Below roughly half a pawn, most positions are simply unclear, and the difference between +0.20 and +0.35 is usually noise rather than insight. Treat small numbers as "balanced" and pay attention to the changes instead.

Mate scores

When the engine sees a forced mate it stops counting pawns and counts moves instead:

A forced mate outranks any material count. M5 is better than being nine pawns up, because it ends the game. This is also why the move classifications treat mate separately: see how moves are classified.

Why the bar stops moving

The bar fills from White's end and saturates at ten pawns in either direction. An evaluation of +10.00 and one of +25.00 look identical, and any forced mate pins the bar completely to one end.

That is deliberate. Past about ten pawns the position is decided, and a bar that kept scaling would spend its whole range on the difference between "winning" and "winning more", leaving no visible movement in the range where games are actually decided. Clamping keeps the interesting band — roughly minus three to plus three — occupying most of the bar.

Before the engine has evaluated a position, the bar sits at an even fifty-fifty in grey rather than guessing. And if you flip the board, the bar flips with it, so your own side is always on your end.

Captured pieces and the material count

Below the board, each side's captures are listed with a +X next to whoever is ahead. The values are the standard ones:

The king has no value here — it is never captured, so counting it would just add a constant to both sides.

One detail worth knowing: the advantage is computed from the material currently on the board, not by adding up the captured pile. Those two numbers come apart the moment a pawn promotes. If you promote to a queen, nothing was captured, but you are genuinely eight points better off than you were — counting captures would miss that entirely, while counting the board catches it.

Material and evaluation disagree, and that is the point

The material count and the engine evaluation measure different things. Being a pawn up with a shattered king position is +1 on material and possibly -2.00 to the engine. When the two disagree sharply, that gap is the story of the position: someone has traded material for activity, and the engine is telling you whether the trade was sound.

Watch changes, not values

The most practical habit is to stop reading the evaluation as a verdict and start reading it as a derivative. A position sitting at +0.400 for twenty moves tells you very little. A position that moves from +0.30 to -1.40 in one move tells you exactly where to look, and the move list will already have coloured that move for you.