Was the Immortal Game sound? What Stockfish says, move by move

Magnus Carlsen lost a five-minute game on 30 September to Volen Dyulgerov, who streams as Witty Alien. ChessBase India's account calls Carlsen's position winning and names 20.Rde1 as the decisive error. Dexerto reports that the game was blitz with no increment and that Carlsen resigned after his 21st move.

A world number one blundering at speed makes news because it is rare. Mistakes in famous games are not new, though. ChessBase's retelling of the Immortal Game says engines find it full of them. It was a casual game played on the side of the London 1851 tournament, on 21 June, and Anderssen, as White, won it by giving up both rooks and his queen.

Kieseritzky, with Black, had the better of the opening. After 10.g4 the engine has him at about 60% to win, his high point. He never gets higher, and from move 11 he is worse for the rest of the game.

Anderssen (White)

86.2%

Kieseritzky (Black)

77.4%

By our accuracy measure that is 86.2% for Anderssen against 77.4% for Kieseritzky. It is a decided game with errors on both sides, not a perfect attack against a perfect defence.

11...cxb5 takes a bishop, which looks natural. Wikipedia calls it Black's critical mistake and the engine agrees: it costs about 17 points of win chance, and the engine would have played the quiet 11...h5. Black goes from slightly better to about 1.4 pawns worse.

16...Bc5 costs another 12 points (the engine prefers 16...Qc6) and White is now about 4 pawns better by the evaluation, even though Black is two points up in material. The reason is in the pieces. After fifteen moves every piece on Black's back rank except the queen was still on its starting square, and the queen had moved six times.

Here Black can capture either rook. 18...Qxa1+ is the engine's choice and leaves White about 1.5 pawns better. 18...Bxg1 takes the other rook and gives White about 3. Two moves later 20...Na6 allows a forced mate in three. The only way to stay alive was 20...Ba6, and even that leaves White around 3 pawns up.

The famous ones came after the game was already going White's way. 18.Bd6 was not the engine's first choice (18.d4 was), and it cost White about 12 points of win chance, though White stayed better. The queen sacrifice 22.Qf6+ is the second move of a three-move mate that existed from the moment Black played 20...Na6.

That does not make the game less beautiful. It makes it a better teaching example than the legend: development and open lines were worth more than the material, and the side that kept its king safe and its pieces working won.

Pick a game you were proud of and run it through an engine. Then look at the move you remember best and ask whether a quieter move earlier was the one that decided the game.

How this was measured: the same Stockfish 18 build this site's analyser runs, at depth 16 on every move and depth 24 on the 24 moves that lost the most ground. Scores are in pawns from White's side. Another engine, depth or computer will move individual numbers a little; the shape of the game stays the same.