FlawChess
Engines are flawless,humans play FlawChess.
FlawChess
Engines are flawless,We took every game in our benchmark database where a player reached the middlegame with a clear advantage (about two pawns up by Stockfish's count) and checked how those games actually ended. The answer should change how you think about "winning positions".
All 117,312 games that entered the middlegame with a ≥ two-pawn engine advantage, all ratings combined.
Players who start the middlegame two pawns up go on to lose roughly one game in four. And this is not just a beginner problem: even at master level (2400 Lichess), nearly one of these games in five still ends in defeat for the player who had the advantage.
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 12%. Strong players simply don't hand over (or receive) free material in the opening.
Split every one of these games into four endings and something surprising appears: the share of sustained-lead wins (the lead never dips below two pawns and converts) barely moves with rating: about 43% everywhere. What improves with strength is everything else: winning anyway after the lead slips, and above all, not losing outright.
The same advantage survives very differently depending on the clock. At every rating, leads die most often in blitz and least often in classical. The gap is brutal at the top: a 2400 loses a two-pawn lead almost twice as often in blitz (22%) as in classical (12%).
Here's the flip side. Among games where the advantage never dipped below two pawns for even one move of the middlegame and endgame, the leader converted 95–99% of the time. The starting size of your lead matters far less than its journey: protect it move by move and the point is essentially banked.