Converting a winning position is a separate skill

The short answer

You convert by making the position simpler, not by making it better. Trade pieces, kill counterplay, give material back when it buys you a position with nothing left to calculate, and check every check and capture your opponent has before you touch a piece.

Getting a winning position and converting one reward opposite instincts. To get the advantage you looked for the sharpest move. To keep it you want the dullest move that still wins. Most players never switch modes, which is why the same person can outplay someone for 25 moves and then hand it back in three.

It goes wrong before the endgame

Across 179,928 engine-classified moves from 5,754 games analysed on Mated, 9.1% of middlegame moves are a mistake or worse, against 2.3% in the endgame. That is a four-fold difference in the same games by the same players.

This is not a claim that endgame technique is easy. It says something narrower and more useful: by the time you reach an endgame the position has usually stopped being sharp enough to lose in one move. The dangerous stretch is the ten or fifteen moves after you win the material, while queens are still on and your opponent still has a piece pointing at your king.

So the standard advice — "study endgames, that is where won games are lost" — is aimed at the wrong part of the game for most club players. You do not lose the rook endgame. You lose the position that was going to become a rook endgame.

The damage is concentrated, not spread

In that same set of 179,928 moves, the average move gives away 95 centipawns against the engine's best, but the median is far lower — a small number of very bad moves carry most of the loss. And 3.8% of moves are outright blunders with another 4.2% mistakes, so roughly one move in twelve is a real error before you count the 18.5% that are inaccuracies.

That shape matters for conversion. If your losses were spread evenly across every move, you would need to play better chess in general. They are not. One move in a won game does almost all of the damage, and it is usually a move you played in under ten seconds because you thought the game was over.

The practical version: you are not trying to raise your average move quality when you are winning. You are trying to remove the tail. Those are different projects. Removing the tail is mostly a checking habit, not a knowledge problem.

Give the material back

You are up a rook. Your opponent's knight lands on a square that forks your queen and a rook, and there is one move that keeps everything — you can see it takes some calculation, and there is a follow-up you have not looked at yet.

Take the knight with the rook. You are still up a piece, the fork does not exist, and the position now contains nothing you can get wrong. If the engine says the greedy move is +7.2 and the simplifying capture is +4.1, take the +4.1. Nobody has ever lost from +4.1 by accident. Plenty of people lose from +7.2 by missing move four of a five-move line at 40 seconds on the clock.

This is the single habit that changes results fastest, in my experience of watching club games. It feels wrong because giving back material feels like losing something. What you are buying is the removal of your own error rate from the position.

Trade pieces, keep pawns

When you are up material, every trade of a pair of pieces increases your relative advantage and reduces the number of things that can attack your king. Pawn trades do the opposite job — they open lines and, in the extreme, take you towards drawn endgames.

The order that matters most is queens first. A queen is the piece that generates counterplay out of nothing: perpetuals, back-rank tricks, forks on two loose pieces. Once queens are off, a piece up at club level is close to automatic.

Two exceptions worth knowing before you simplify:

• Opposite-coloured bishops with few pawns. A pawn or even two up can be a dead draw. Check what the final structure looks like before you trade into it.
• Rook's pawn plus the wrong-coloured bishop. Same warning. If your h-pawn and your bishop do not control h8, the endgame is drawn no matter how long you push.
• Everything else: trade. Rook endgames a pawn up are drawish in theory and won in practice at club level, because your opponent has to defend accurately for thirty moves.

Assume every opponent move is a threat

A player in a lost position plays differently. They stop trying to find the objectively best move and start setting traps, because a trap is their only remaining path to half a point. That is correct play from them, and it means the base rate of tricks in their moves goes up exactly when your attention goes down.

So change what you do between moves. When you are winning, before every move, list your opponent's checks, captures and threats to your king and to any undefended piece. Out loud in your head, every time, even when the answer is obviously nothing. It costs fifteen seconds. The data above says the move that loses your game is a single move, so the whole job is catching that one.

If you want to know which of these actually costs you points, the answer is in your own games rather than in an article. Filter your last fifty for the ones where the evaluation was above +3 for you and the game did not end in a win, then look at the move where it swung. Mated does this by running Stockfish over every position from your Chess.com or Lichess history and grouping the collapses, but you can do the same by hand with a free analysis board and an afternoon. What you want is the pattern: back rank, hanging piece on the queenside, king walk in a queen endgame, time.

What nobody has measured well

There is no good public number for how often club players throw away won positions, because "winning position" has no agreed definition. Is it +2? +3? Winning for a 2000 but not for a 1200? Everything you read on this, including the parts of this article that are not tied to the move data above, is an estimate.

What can be said from the 179,928 moves is narrower and still worth having: errors cluster in the middlegame, and a handful of moves account for most of the evaluation lost. Whatever the true conversion rate is, the fix points in the same direction.

One thing to add that is not about technique. Look at your clock usage in a game you won and then lost. My guess — and it is a guess, not something these numbers show — is that you spend your time in the phase where you are building the advantage and almost none in the phase where you are protecting it. The evaluation swing is the moment to slow down, not the moment to relax. If you finish a won game with eight minutes unused, you did not save time, you spent it on the wrong half of the game.