The Chess Eval Bar, Explained: What It Measures and When to Trust It

The chess eval bar shows a single number: the engine's judgment of the position, measured in hundredths of a pawn and reported from White's point of view. +1.20 means the engine's search concluded White stands better by about a pawn's worth of position — not that White is up a pawn, and not a promise about the result.

What is the eval bar showing?

The bar is a graph of one value, the engine evaluation, expressed in pawn units. Everything on the board — material, king safety, piece activity, pawn structure, concrete tactics — gets compressed by the engine's search into that one number. The pawn is just the unit of account: +0.80 with equal material means the engine thinks White's positional advantages are worth roughly four-fifths of a pawn.

Two conventions matter for reading it. First, sign: the bar shows the number from White's point of view by convention, so a negative number is good news for Black. Under the hood, the engine protocol actually reports scores from the side to move — every tool converts before displaying — which is why in AxiomChess every stored evaluation is labeled with whose point of view it carries: a silent sign flip corrupts every judgment built on top. Second, scale: the bar's visual fill is usually a squashed version of the number, so the distance from +1 to +3 looks bigger than from +6 to +8. The number is the datum; the bar is the illustration.

How is the chess eval bar calculated?

The number comes from a search, and the search has a budget — which is why the same position can show different evaluations in different tools. The engine explores the tree of possible continuations, evaluates the resulting positions, and reports the value of the best line it found. Give it more time or depth and the answer can change, because a deeper search sees consequences the shallower one missed.

That budget is exactly what a trustworthy tool should disclose. For every game analyzed on AxiomChess, each position is evaluated by Stockfish 18 in a single bounded pass (depth 20, with an 8-million-node ceiling), running in your own browser, with no account needed. Disclosed on purpose: an evaluation you cannot reproduce is an evaluation you cannot argue with. You can watch the same engine work on any position on the free analysis board.

The eval bar and your accuracy score are related but distinct: the bar judges the position in front of you, while accuracy summarizes how much winning chance your moves preserved across a whole game — that construction is covered in the accuracy guide.

How do centipawns turn into winning chances?

Through a fixed conversion curve — and knowing its shape changes how you read the bar. The relationship is not linear: through the published curve, the first pawn of advantage is worth roughly nine points of win chance, while the step from +4 to +5 adds about five. Around 0.00 the curve is steepest, so small swings matter most; by +5 you are near 86%, and each further pawn buys less. Lichess publishes the logistic curve it uses for exactly this conversion on its accuracy methodology page. The version AxiomChess ships even clamps evaluations at ten pawns before converting — past that, the percentage stops moving at all, because as a probability question the game is already answered.

AxiomChess shows both readouts — the centipawn number and the winning-chance percentage derived from it. One is computed from the other through that fixed curve, which means the two can never disagree; the percentage is a translation, not a second opinion. The translation exists because pawn units answer the engine's question ("how big is the advantage?") while percentages answer the player's ("how often does this win?"), and mid-game decisions are made with the second question.

Why does the eval bar jump around?

Because the search revises itself as it deepens, and because some positions genuinely hinge on one move. Engines search iteratively — the reported number is the current best estimate and can move when a deeper iteration uncovers a resource the previous one missed. In sharp tactical positions the evaluation can legitimately swing several pawns between adjacent moves, because the position's truth really did change when a threat appeared or a defense ran out.

There is also a boundary effect: an engine can only report what its search reached. A fortress it cannot break, or a plan that pays off beyond its depth budget, gets evaluated on what is visible inside the search. This is the other reason two sites can disagree about the same position — different engine versions, depths, and settings are different searches, and each bar honestly reports its own.

Two AxiomChess design choices exist precisely because of this. Score lines the engine flags as mere search bounds are rejected by the parser, so a stopped search can never publish its half-finished estimate as the final eval. And the bounded pass — same depth, same node ceiling every time — means re-analyzing the same game shows the same bar: the number is allowed to jump for chess reasons, never for budget ones.

When should you not trust the eval bar?

Trust the bar about the position; be skeptical of what it implies about your game. Four cases come up constantly:

• 0.00 is not a handshake. It means the engine found balance with best play from both sides. Between humans, one side's balance may be effortless and the other's a tightrope — equal chances for the engine, not for you.
• A big number is not a plan. +3 tells you a win exists somewhere in the tree; it does not tell you the idea. If you cannot name the plan, the bar's confidence is not yours to borrow.
• The bar has no memory and no nerves. It re-judges each position cold, which is precisely why the bar can stay calm while a player collapses — and why conversion is its own skill with its own leak.
• Below a floor, stop arguing with it. Deep in a lost position, differences between defenses stop being meaningful in practical terms. The AxiomChess coach applies this to itself: it stays silent once the side to move is below 20% winning chances, because coaching the arrangement of deck chairs is noise.

The bar is the best single number chess has. It is still one number — the judgment of one search at one depth — and reading it well means knowing which question it answers.

FAQ

What does +1 mean on the eval bar?

The engine judges White better by the equivalent of one pawn — as a combination of material and position, not necessarily an extra pawn on the board. Through the published Lichess curve, one pawn of advantage corresponds to modestly favorable winning chances, roughly 59% — a lead, not a result.

Is 0.00 on the eval bar always a draw?

No. It means the engine's search found balance with best play — which can be a dead-equal endgame or a razor-sharp position where one wrong step loses for either side. Between club players, 0.00 positions are decided all the time. The bar reports the position's balance, not the game's likely outcome.

Does the eval bar show forced mate?

Yes — engines switch from pawn units to a mate count, shown as M5 or #5 rather than a pawn-unit evaluation. A mate score is a different kind of claim: not "better by this much" but "forced win in this many moves." AxiomChess renders mate lines with the mate notation and shows 100% winning chances rather than pretending a centipawn number that large means anything.

Why do different sites show different evals for the same position?

Because an evaluation is the output of one search, and sites run different searches — different engine versions, depths, node budgets, and settings. Both numbers are honest reports of their own search. This is why tools should disclose engine and depth; a number without its budget cannot be compared to anything.

Can you use an eval bar during a rated game?

No — engine assistance during rated play is cheating on every major platform (Lichess bans it explicitly in its terms of service, as does Chess.com's fair-play policy), and fair-play systems are built to detect it. The eval bar belongs to analysis after the game and to watching broadcasts, where it turns grandmaster chaos into a story spectators can follow.

The eval bar tells you how big the advantage was; the useful question is which of your moves gave it away, and whether the same kind of move does it every game. Analyze your games free — every evaluation disclosed, depth and all.