Why the Engine Hates a Move You Like
You play a natural developing move and the evaluation drops half a pawn. Four things the engine sees that a club player usually does not, and how to check which one applies.
· 3 min read
You develop a knight to a square that looks obviously correct, the evaluation drops from +0.4 to −0.2, and nothing about the position appears to have changed. This is one of the most common experiences people have with an engine, and it is worth knowing what is usually behind it.
The number is not the explanation. The continuation is. Almost every time, the reason is visible within five moves of the line the engine gives after your move, and the four causes below cover the overwhelming majority of cases.
A tactic you have not reached yet
The most frequent reason, and the easiest to verify. Your move loses material to a sequence starting three or four moves later, often after a quiet intermediate move that does not look like part of a combination.
Play the engine's line out on the board rather than reading it. The tactic will be at the end, and it is almost always a pattern you already know in isolation — a fork, a pin, a back-rank problem — appearing in a position where you were not looking for it.
A square handed over permanently
Pawn moves cannot be taken back, and the engine weighs that heavily. Playing a pawn to a square that looks aggressive frequently concedes the square in front of or beside it for the rest of the game.
The tell is that the evaluation does not recover. A tactical problem resolves once the tactic is dealt with. A structural concession stays in the number, drifting slowly worse over the following ten moves as the opponent's pieces take up residence on the square you gave away.
Your king, later
King safety is one of the largest terms in an engine's evaluation, and it is assessed well before there is any attack. Moving a pawn in front of your own king, or delaying castling by one move to grab something, often costs more than the thing you gained.
This one is genuinely hard to see as a human, because at the moment you play the move there is no attack to point at. The engine is pricing in an attack that does not exist yet against a defence that has just got worse.
The move is fine and you are being fussy
Worth checking before assuming a deep reason. If your move reads +0.31 and the engine's reads +0.42, nothing is wrong with your move. That difference is below the level at which an engine's ranking means anything.
The threshold worth caring about is around half a pawn. Below that, the engine has expressed a mild preference, not a judgement. Above it, something specific has happened and one of the first three causes will explain it.
| Cause | Tell in the line | Does it recover? |
|---|---|---|
| Tactic | Ends in a capture or fork you did not see | Yes, once dealt with |
| Square conceded | No captures at all, pieces simply improve | No, it drifts worse |
| King safety | Moves converge on your king | Only if you spend time repairing it |
| Nothing wrong | Gap under 0.3 | There is nothing to recover from |
How to find out which
Set the position up before your move, get the engine's recommendation, then play your move instead and analyse again. Two numbers and two lines.
The size of the drop tells you the severity. The continuation after your move tells you the mechanism. If the line contains a capture you did not anticipate, it is a tactic. If it contains no captures at all and the pieces simply improve, it is structural. If the moves converge on your king, it is safety.
That takes about a minute per move, and it is the single most useful thing you can do with an engine on a game you have already played.
Common questions
- Why does Stockfish dislike a move that looks natural?
- Most often because it creates a weakness that only matters later, allows a tactic three or four moves away, or hands over a square the opponent's pieces want. Play the engine's continuation out and the reason usually appears within five moves.
- Can the engine be wrong about a move?
- At shallow depth, regularly. At depth 18 or more in a normal middlegame, almost never in the sense of being mistaken about the value. It can still recommend a move that is impractical for a human, which is a different thing from being wrong.
Try it on your own position
The analysis board runs the same engine in your browser, free. The app adds what a web page cannot: reading a position straight from a photograph, saved boards, and analysis with no connection.
