The Eval Bar, and What It Hides

The bar is a single number drawn as a picture. Everything about how it behaves follows from that, including the ways it misleads people watching a game.

It is compressed, not linear

A linear bar would be useless. An advantage of ten pawns and one of twenty are both completely winning, and a bar that distinguished them would spend most of its length on differences nobody cares about while squashing the interesting range near equality into a few pixels.

So the mapping is compressed. Roughly:

The practical consequence is that the bar is most sensitive exactly where you want it: near equality, where small changes matter. It stops telling you much once someone is clearly winning, which is fine, because at that point there is nothing left to distinguish.

What it hides

Whether the position is easy to play. A +1.5 requiring three precise moves and a +1.5 where every reasonable move keeps the advantage draw an identical bar. For a human, those are entirely different positions.

How stable the number is. A bar sitting at +0.8 after a settled depth-30 search and one at +0.8 from a depth-14 search that is still moving look the same. Broadcast bars usually run shallow, which is why they sometimes lurch for no visible reason.

Time on the clock. A player with a winning position and thirty seconds is not winning in any practical sense. The bar has no opinion about this and it is frequently the most important thing on the board.

Which side finds it easier. Compensation for a sacrificed piece often means one side has a clear plan and the other must defend accurately for twenty moves. Equal on the bar, very unequal to play.

Where it is genuinely useful

Watching the shape of a game. A bar that stays near the middle for forty moves and then moves decisively tells you the game was balanced and then was not, and where. That is a real summary and it is faster to read than any list of moves.

It is also good for finding the moment to study. When reviewing your own game, the point where the bar left equality is usually the position worth understanding, and it is frequently several moves before the mistake you remember making.

How to read it during a broadcast

Two habits make it much more informative. Look at the depth if it is shown, since a shallow bar is a guess. And watch for the bar recovering on its own without a move being played, which means the engine is still thinking and its first answer was provisional.

Beyond that, treat the bar as a picture of the position's value and nothing else. It does not know who is tired, who is in time trouble, or which side has to find the only move. Those decide games too.

Mate scores break the scale

When a forced mate appears, the bar goes fully to one side and stops. There is nowhere further for it to go, and no distinction is drawn between a mate in two and a mate in twenty.

That is correct as far as it goes, but it removes information at exactly the moment things get interesting. A mate in twenty-five in an endgame is a very different practical situation from a mate in two, particularly with the fifty-move rule in play. The bar cannot show that, so the numeric score beside it is the thing to read once the bar has bottomed out.

Watching your own games back

The bar is most useful in review, played back move by move rather than read at a single moment.

Three shapes come up repeatedly:

• A cliff. One move, large drop. A concrete blunder, and the easiest kind to learn from.
• A slope. Steady drift across ten or fifteen moves with no single bad move. Usually a plan that was wrong from the start, and much more valuable to understand than a cliff.
• A sawtooth. The bar swinging back and forth. Both sides missing things in a sharp position, which frequently means the position was beyond both players and the opening choice is what to examine.

The slope is the one people miss, because no individual move looks bad enough to flag. It is also where most rating points are lost.