Read the sign before the size
Quick answer: A chess engine evaluation is its current estimate of the position, with positive values favoring White, negative values favoring Black, and values near zero suggesting relative balance. The score isn't a promise about the result, and it can change as the engine searches further or the position changes. Use it to locate a decision worth explaining, then translate the engine line into a board reason you can understand.
The sign comes first. Before worrying about the size, check which side the engine favors. That simple habit prevents a surprising amount of confusion when you flip the board or review a game from Black's point of view.
Then pause.
A number can't tell you whether the preferred move was easy to find, whether your alternative was practical, or what you believed during the game. It reports the engine's assessment under its current search. Your review still has to explain the position.
This is why chasing a prettier score can become empty. If you don't know which piece improved, which threat disappeared, or which line failed, the evaluation has named a change without teaching you anything about it.
What the evaluation number represents
I read the official Chess.com guide to computer evaluation numbers before writing this explanation. The guide says an evaluation is an engine's assessment of the current position and describes the advantage in pawn units.
Its positive example is `+2.25`, which it uses to indicate that White has the perceived advantage. Negative values favor Black, while values close to zero suggest a relatively even position. Those are sourced examples, not conversion rules for player ratings.
Think position, not scoreboard.
The unit offers a familiar scale for the engine's judgment, but the board may not contain a matching material difference. The same official guide says activity, king safety, pawn structure, and other positional factors can affect the evaluation.
So don't read a positive score as White has collected that many pawns. Read it as the engine currently prefers White's position by the displayed amount on its evaluation scale. Then inspect the line to learn what supports that preference.
The engine comparison guide discusses engine choices. This article has a different job: helping you read the output without turning it into a false certainty.
Material is only part of the story
Suppose the material count is level but the evaluation favors a side. Don't assume the engine is broken. Start looking for features that change what the pieces can actually do.
Check king safety. Check activity. Inspect passed pawns, weak squares, trapped pieces, overloaded defenders, and forcing moves. These are candidate explanations, not a checklist that proves the score. The engine line decides which feature matters in the position.
Follow the preferred move and the strongest reply. If the advantage survives because a rook enters an open file, name that access. If it survives because a threat forces a passive defense, name the threat and the piece tied to stopping it.
Keep the language modest.
You aren't trying to produce a grandmaster annotation from a single engine line. You're trying to connect a numerical judgment to a board fact that would change your next decision.
Sometimes the explanation is tactical. Sometimes it is a sequence of quiet restrictions. If you can't explain the score yet, save the position and continue the line carefully rather than inventing a strategic slogan that sounds wise.
Depth explains why the score can move
I checked Chess.com's official analysis depth guide for its current definition. It says the depth display describes how many moves ahead the engine is looking while deciding on preferred moves and move labels.
More search can reveal a reply that wasn't visible earlier. That is why a quick evaluation may shift as analysis continues, a limitation also stated in the evaluation guide. The early score wasn't a contract. It was the current output.
Let the position settle before making a strong claim.
This doesn't mean you should worship the largest depth available or wait forever. For training, the useful question is whether the line has become stable enough to explain the decision you made. If the leading move and evaluation keep changing, mark the position as unsettled and avoid writing a confident lesson.
Depth also isn't the same as understanding. A deeper engine result can be more settled while your explanation remains weak. You still need to identify the move that changes the board and the reply that makes your candidate fail.
When two screens disagree, check their perspective, engine settings, and search state before claiming the products evaluate chess differently. You haven't run a controlled engine comparison merely by opening the same game in separate tabs.
A score swing marks a question
A sudden evaluation change is useful because it points toward a decision worth reviewing. It doesn't finish the review. The move may lose material, allow a forcing attack, abandon a defense, or miss a chance that existed only briefly.
Hide the answer again.
Return to the position before the swing and write what you thought was happening. List your candidate and the opponent's forcing replies. Then reveal the engine move and find the earliest point where your line stops matching the engine's.
That mismatch is the lesson source. If you missed a check, the repair is calculation or threat scanning. If you saw the check but misjudged the ending, the repair is evaluation. If you understood the position and moved too quickly, the repair is a decision habit.
The same score change can therefore produce different training tasks for different players, because the number doesn't contain the reason behind the human choice.
Use the Game Review workflow when labels are part of your screen, but don't let the label replace this diagnosis. A colorful category describes the move outcome. Your note should describe the thinking gap.
Read the line as a sequence of reasons
Engine lines look authoritative because every move is already chosen. Resist copying them wholesale. Cover the continuation and predict the reply after each candidate.
Ask why that reply is forcing.
Maybe it gives check, attacks a queen, creates a promotion threat, or removes a defender. Maybe it isn't forcing at all, and the engine simply prefers a quieter improvement. Your explanation should match what the line demonstrates, not what you hoped the move meant.
Now compare nearby alternatives. If several moves preserve a similar evaluation, the position may offer practical choice rather than a single magical answer. If your move alone causes the score to collapse, isolate the board change that makes it different.
Don't claim a move is unplayable merely because another move scores better. Look at the size and stability of the difference, the resulting line, and whether you can state a concrete consequence. The engine's preference is evidence, but your wording should stay within that evidence.
A useful annotation sounds like this: moving the knight removed the only defender of the pawn, so the forcing capture worked. It doesn't sound like this: the position became strategically bad. The second sentence hides the mechanism.
Winning evaluations still need conversion
A favorable score doesn't move the pieces for you. The official evaluation guide warns that an advantage doesn't guarantee a win, especially in human play. Treat that warning seriously during training.
Search for the plan behind the number.
Does the engine want a trade, a safer king, an active rook, or a passed pawn pushed? Follow the line until you can name the purpose, then return to the starting position and choose the move without seeing it.
If the score stays favorable across several moves, compare plans rather than hunting for a single sequence to memorize. A stable advantage may leave room for choices, while a tactical position may demand precision. The line will show which kind of task you face.
The winning position guide focuses on practical conversion. Engine evaluation belongs at the diagnosis stage. It tells you that an opportunity may exist, while the board still requires a plan and accurate replies.
Don't celebrate the number. Explain the move. Then ask what the opponent can try to make the plan difficult. That question keeps the analysis connected to an actual game rather than an empty advantage bar.
Write notes the engine cannot write for you
End the analysis pass with a note about your decision, not a pasted variation. Record the position, your original candidate, the reply you missed, and a cue that would have helped you look in the right place.
Keep it specific.
Before moving this defender, check what it is protecting is useful. Calculate better is not. Your future self needs a visible trigger, not a scolding.
Save the engine line as evidence underneath the note, then hide it when you revisit the position. Try to recover the reason before the move. If you remember the move but can't explain the threat, the note still needs work.
Compare your explanation with the board after the position changes. A cue that depends on the exact arrangement may be a memorized answer. A cue that points to a recurring relationship, such as an overloaded defender or exposed line, can be tested in other positions.
Keep the position active until you can state the sign, identify the favored side, explain the decisive board fact, and choose a sensible continuation without borrowing the engine's voice.
One thing worth knowing before you build a routine on this. A free Chess.com account can use Game Review and still see accuracy, move classifications and Self Analysis. Unlimited Game Review and some of the deeper analysis options need a Premium membership, and Diamond adds Cloud Analysis with the Deep and Maximum engine settings. Chess.com does not publish an exact free daily allowance, so check what your own account offers before planning around it.
Lichess has no paid tier at all, which is the practical reason a lot of players do the heavy review work there and keep Chess.com for playing.