How to read an eval graph without lying to yourself
The eval graph is the first thing everyone looks at after a game and the thing they take the least away from. Usually the takeaway is “there’s the blunder”, which was already obvious, followed by closing the tab.
There is considerably more in it than that, once you know which shapes to look for.
First: it is not measuring what you think
The line is the engine’s evaluation of the position. It is not a measure of how well you played, and it is not a prediction of the result.
Two specific consequences:
A flat line at +0.3 for forty moves is not a boring game. It often means both players navigated something genuinely sharp without error. The graph cannot show effort.
A drop is not automatically a mistake you could have avoided. Sometimes the position was already lost and the engine only noticed three moves later. The graph shows when the evaluation changed, not when the game was decided.
This is why we draw the graph in win probability rather than raw centipawns. A swing from +0.2 to +1.4 near equality changes your winning chances far more than a swing from +6.0 to +7.2, and a centipawn axis makes the second one look bigger. It is not.
The four shapes
The cliff
Flat, then a single vertical drop, then flat again at the new level.
This is the one everyone finds, and it is genuinely the simplest to learn from: one move, one identifiable error. Go to that ply, look at what the engine wanted, and understand the difference.
The useful question is not “what was the best move” but “what would I have had to notice?” A cliff caused by missing a knight fork is a different problem from one caused by a positional misjudgement, and they need different training.
The staircase
A series of small drops in the same direction, none of them individually alarming.
This is the most instructive shape and the one people skip, because there is no dramatic moment to point at. Nothing here looks like a blunder. You lost the game in eight moves of slightly-worse, each of which felt reasonable.
Staircases usually mean a plan problem, not a tactics problem. You were executing something that does not work in that structure. Find where the staircase starts — not where it ends — and ask what you were trying to do.
The sawtooth
Up and down repeatedly, both players trading errors.
Almost always a time-pressure or complexity signature. Check the clock readings against the swings. If the sawtooth starts when both clocks drop under a minute, the graph is telling you about the clock, not the position.
Sawteeth are the least useful games to review move by move and the most useful to count. If most of your games look like this, your problem is upstream of any individual move.
The wall
A sharp move to a large advantage that then holds perfectly flat to the end.
Worth reviewing even though you won. The flat part is the skill: converting is a separate ability from getting the advantage, and most players below 1800 are much better at the first than the second. If your walls tend to wobble before the finish, conversion technique is a cheaper source of rating points than anything in your opening.
The trap
Here it is: the eval graph rewards the winner and punishes the loser, regardless of who played better.
If your opponent blunders on move 12 and you convert cleanly, the graph shows a beautiful step and a flat line. It looks like an excellent game. You may have played thirty completely unremarkable moves in a position where nothing could go wrong.
If you play a genuinely difficult game against a stronger opponent, defend well, and lose to one late slip, the graph shows a slow decline and a drop. It looks bad. It might have been your best game of the month.
Reviewing only your losses, guided by the shape of the graph, teaches you to avoid difficult positions. That is the opposite of improving.
How to actually use it
- Find the largest swing and go there first. One move usually accounts for a disproportionate share of the damage. That is what the key moments list is for.
- Then look for a staircase. The slow decline is where the repeatable mistake lives.
- Check the clock at every swing. If the swings cluster in time trouble, stop reviewing the moves and go read the clock-pressure chart instead.
- Review a win. At least one in three. Wins are where the “I got away with it” moves hide, and they are the ones you will repeat.