The opening mistakes club players actually make

Every chess book tells you which moves are good. Almost none of them tell you which bad moves people actually play, and those are two very different lists. An engine is no help here either: every move Stockfish considers is already a reasonable one, so it will never tell you that a third of the club players you face will walk into the same mistake on move six.

So we did it the other way round. For each of the 61repertoires in the ChessReflex library we pulled every position the repertoire reaches with the opponent to move, asked the Lichess opening explorer what players rated 1600–2200 actually play there, and then asked Stockfish what each of those choices is worth. The interesting moves are the ones in the overlap: played often, and simply bad.

53 of the openings we checked contain one. On average the losing move is played by 13% of players in that position, and 2of them do not merely lose material — they lose by force.

Why popularity is the half nobody measures

Consider what it takes for a mistake to matter to you. It has to be bad, obviously. But it also has to be a mistake somebody makes, and it has to be a mistake somebody makes against the openings you play. A refutation of a line nobody enters is a party trick.

This is why engine analysis on its own produces such unusable opening advice. Point Stockfish at a position and it hands you a principal variation of moves that are all excellent, played by people who are all excellent. The moves that will decide your next hundred games are nowhere in that output, because they are beneath the engine’s notice. You need a database of ordinary players to find them, and an engine to price them.

The list

Each row is the most-played losing move in that opening, as measured in the position named. The cost is what Stockfish makes of it against best play, confirmed at depth 26 — shallow searches call all sorts of ordinary moves refutations, so nothing here went in on a first opinion.

What the pattern tells you

Read down the list and the same few errors keep reappearing in different costumes.

1. The automatic move

A striking share of these are moves that are correct one move earlier or one move later. h3 to question a bishop, Nf6 to develop, a recapture that looks forced. They are played quickly because they look like the move you always play in that structure, and the reason they lose is almost always a concrete tactical detail that the general principle does not know about.

2. The greedy move

Taking a gambit pawn a second time, or grabbing a b-pawn with the queen. These appear at eye-catching frequencies because the material is genuinely there and the punishment is genuinely several moves away. If you play a gambit, this is your entire edge: your opponents will take, and then take again.

3. The early queen

Queen sorties that win a pawn or give a check and lose two tempi. The engine’s numbers on these are often modest — half a pawn, a pawn — which understates them badly, because the resulting positions are much easier to play from one side than the other.

The most expensive single move in our sample is Nxe5 in the Caro-Kann: White's Advance, after 1.e4 c6 2.d4 d5 3.e5 c5 4.Nf3 Nc6 5.dxc5 Bg4 6.c3. It appears in 19% of games at this level and costs 5.8 pawns — and the move that holds, e6, is not a difficult move to find once you know to look.

How to use this

The wrong lesson is to memorise the table. The list changes as the database changes, and knowing that a specific move is bad in a specific position is worth almost nothing if you never reach the position.

The useful lesson is narrower and more durable: in every opening you play, there are two or three places where your opponent has a real choice, and one of those choices is usually much worse than it looks. Those are the positions worth knowing cold. Everything between them is filler you can work out at the board.

That is also the reason ChessReflex drills branch points rather than whole lines. Across the whole library, the openings contain 1,156 genuine decisions and 18,685 moves — the decisions are the part that decides games, and they are about a sixteenth of the work.

If you want the version of this specific to what you play, every opening page carries its own figures: what your opponents choose in each position, how they score with it, and which of their choices is the mistake. Browse the library.

Method

Move frequencies come from the Lichess opening explorer, restricted to the 1600, 1800, 2000 and 2200 rating bands across blitz, rapid and classical. Positions with fewer than 400 games in the sample were discarded. Evaluations come from Stockfish: candidates were screened at depth 20 and every published claim was re-confirmed at depth 26, comparing the move against the engine’s own best in the same position. A move qualifies as a mistake here when it costs at least 1.5 pawns. Nothing on this page was written from memory or from a book.