You're two pawns up.
Do you actually win?
In one of our recent exchanges, GM Noël Studer of Next Level Chess suggested looking into what actually becomes of a big engine advantage at the end of the opening. So I had a close look at games in the FlawChess benchmark database where a player, facing an opponent of nearly the same rating, reached the middlegame with a Stockfish evaluation of at least +2.0, the engine's way of saying "about two pawns up". That can mean extra material, a superior position, or both. Then I checked how those games actually ended for different rating levels and time controls. The answer might change how you think about "winning positions".
All 82,023 games that entered the middlegame at +2.0 or better, all ratings combined. "At least" matters: the median lead here is +3.6 pawns, and a bare two-pawn lead does considerably worse than the headline (section 3).
Players who start the middlegame two pawns up or better go on to lose roughly one game in four. And this is not just a beginner problem: even at 2400 Lichess, more than one of these games in five still ends in defeat for the player who had the advantage.
Games LOST by the player who entered the middlegame at least two pawns up
View the data
| Rating | Games with an entry lead | Median lead | Leader loses |
|---|---|---|---|
| 800 | 6,998 | +4.4 | 27.9% |
| 1200 | 23,390 | +4.1 | 27.9% |
| 1600 | 24,962 | +3.6 | 25.2% |
| 2000 | 17,469 | +3.1 | 22.6% |
| 2400 | 9,204 | +2.9 | 21.8% |
At 800, a majority of games (58%) already have a two-pawn advantage for someone when the middlegame starts: openings decide games early. By 2400 that collapses to 10%. Strong players simply don't hand over (or receive) free material in the opening.
Share of ALL games that begin the middlegame with someone two pawns up
View the data
| Rating | Share of games entering with a big lead |
|---|---|
| 800 | 57.5% |
| 1200 | 44.5% |
| 1600 | 30.8% |
| 2000 | 18.5% |
| 2400 | 10.3% |
Two pawns is the floor of this group, not its typical member. Half of these leads are bigger than +3.6 pawns, and roughly one in four is +5 or more, worth about a rook. Size changes everything: a bare two-pawn lead is lost 32% of the time, while a five-pawn lead is lost only 14.6% of the time. The 25% headline is an average across all of them.
Games LOST by the leader, by how big the lead was when the middlegame began
View the data
| Size of the lead | Games | Share of games | Sustained-lead win | Win after slip | Draw | Leader loses |
|---|---|---|---|---|---|---|
| +2 to +3 | 29,561 | 36% | 18.9% | 44.9% | 4.4% | 31.8% |
| +3 to +5 | 33,691 | 41.1% | 43.5% | 27% | 3.6% | 25.9% |
| +5 to +10 | 16,938 | 20.7% | 69.3% | 14.2% | 1.9% | 14.6% |
| +10 or mate | 1,833 | 2.2% | 84.6% | 8.7% | 0.5% | 6.2% |
Split every one of these games into four endings and something odd appears: the share of sustained-lead wins (the advantage stays at two pawns or more the whole way from middlegame entry to the win) barely moves with rating, sitting near 40% everywhere. It is tempting to conclude that masters are no better at holding on to a lead. That conclusion is wrong, and the reason is section 3.
Four outcomes of a two-pawn-or-better middlegame lead
View the data
| Rating | Sustained-lead win | Win after slip | Draw | Leader loses |
|---|---|---|---|---|
| 800 | 40.8% | 28.2% | 3.1% | 27.9% |
| 1200 | 41.1% | 28.5% | 2.5% | 27.9% |
| 1600 | 41.4% | 30.1% | 3.3% | 25.2% |
| 2000 | 40.6% | 32.4% | 4.4% | 22.6% |
| 2400 | 39.3% | 34% | 4.8% | 21.8% |
Weaker players are handed much bigger leads (median +4.4 pawns at 800 against +2.9 at 2400), and bigger leads are easier to hold on to. That mix effect swallows the skill effect whole and flattens the green bar. Compare like with like, the same size of lead at middlegame entry for every rating, and the flatness disappears.
Sustained-lead wins by rating, holding the size of the entry lead fixed
View the data
| Rating | Entry lead +2 to +3 | Entry lead +3 to +5 | Entry lead +5 to +10 | All sizes |
|---|---|---|---|---|
| 800 | 15.6% | 32.2% | 61.5% | 40.8% |
| 1200 | 15.4% | 38% | 66.3% | 41.1% |
| 1600 | 17.9% | 43.7% | 72.4% | 41.4% |
| 2000 | 20.6% | 50.3% | 77.2% | 40.6% |
| 2400 | 23.2% | 53.6% | 78.6% | 39.3% |
At 800, 30% of two-pawn blitz leads end in defeat against 22% in classical, so blitz looks like the worst place to hold an advantage. But split those defeats by how the game actually ended and two very different problems appear. Losing on the board is a rating problem: the three time controls sit almost on top of each other, falling from roughly 20% at club level to 13–16% at 2400. Losing on the clock is a time-control problem: blitz leaders flag away 6–9% of these winning positions at every rating, rapid players about 3%, classical players about 1%, and getting stronger barely moves it. Those flag falls are genuine throwaways: in 41% of them the engine still had the blitz leader two pawns up on the final move.
Games LOST by the leader: on the board vs on the clock
Lost on the board
Lost on the clock
View the data
| Rating | Time control | Lost on the board | Lost on the clock | Disconnect | Total lost | Board only* |
|---|---|---|---|---|---|---|
| 800 | Blitz | 20.1% | 9.4% | 0.6% | 30.1% | 25.6% |
| 800 | Rapid | 22.5% | 2.6% | 1% | 26.1% | 24.7% |
| 800 | Classical | 18.3% | 0.5% | 2.9% | 21.6% | 20.9% |
| 1200 | Blitz | 21.4% | 8.5% | 0.5% | 30.4% | 26.7% |
| 1200 | Rapid | 23.9% | 2.3% | 0.7% | 26.9% | 25.9% |
| 1200 | Classical | 23.5% | 0.6% | 1.4% | 25.4% | 25.1% |
| 1600 | Blitz | 20.3% | 6.2% | 0.4% | 26.9% | 24.3% |
| 1600 | Rapid | 21.9% | 2.4% | 0.6% | 24.9% | 23.7% |
| 1600 | Classical | 21.8% | 1.4% | 1% | 24.2% | 23.5% |
| 2000 | Blitz | 17.4% | 7.9% | 0.2% | 25.5% | 21.5% |
| 2000 | Rapid | 19.1% | 3% | 0.3% | 22.3% | 20.9% |
| 2000 | Classical | 18.1% | 1.2% | 0.5% | 19.8% | 19.2% |
| 2400 | Blitz | 15.6% | 7.4% | 0.2% | 23.2% | 19.2% |
| 2400 | Rapid | 16.3% | 3.5% | 0.2% | 20% | 17.8% |
| 2400 | Classical | 12.9% | 1.4% | 0% | 14.3% | 13.6% |
* Share of losses among games that were actually decided on the board, with clock-decided games removed from the count entirely. The three loss columns are shares of all entry-lead games and add up to "Total lost", to rounding. The chart shows the board and clock components; disconnects (at most 2.2% in any cell) appear only in this table.
What to take to your next game
#- A lead is a promise, not a result. Players who reach the middlegame two pawns up still lose one game in four, and at exactly two pawns the risk is closer to one in three. Keep playing as if the game were undecided, because it is.
- Holding a lead is its own skill. Given the same size of lead at middlegame entry, sustained-lead wins (the lead never dips below two pawns for the rest of the game) climb steadily with rating: 16% at 800 to 23% at 2400 for a two-to-three-pawn entry lead. Winning the material was only half the work; the other half is the unglamorous part, trading down and shutting off counterplay.
- Losing won blitz games? The fix may be a slower time control, not more tactics. Blitz leaders flag away 6–9% of these games at every rating, and getting stronger barely helps. Playing rapid instead cuts that to about 3%, and classical nearly eliminates it.