GM Gukesh D – GM Mahammad Muradli

The opening was uneventful: 7.Qf3 rather than the sharper 7.f4, then 8.Bd2 and a knight that came back via b3 and d4. After 14...Bd7 White stood a quarter of a pawn to the good and had to decide how to open the game. The practical move was 15.f4 or 15.O-O; the database knows them both, and the calculation on moves like these is manageable.

Gukesh thought for eighteen minutes and thirty-three seconds and chose 15.Na4. The knight does not quite attack the queen, because c5 stays occupied by the black pawn, and on the calculation White ended with a small edge. Objectively the move is fine, but the answer Black did not find would have cost the edge: 15...Bxa4 16.Rxa4 d5 17.exd5 Nxd3+ 18.Qxd3, and White has nothing left in hand. The central pawn break is the standard cure for every white piece that is supposed to hold the queenside.

Muradli played 15...Nc4 instead, traded bishops, traded queens, and after 18.cxd3 e5 a clean endgame sat on the board with White still a shade the better side. Gukesh was a whisper richer and eighteen minutes poorer. What you buy with a long think is not always a better position: sometimes it is only the certainty that the position is still playable. Add up the rest of the game before you spend half your clock on one move; the moves after the big think still have to be made.

Up to move thirty neither side had ever given away more than half a pawn. Then Muradli played 31...Bc5, the bishop move almost everyone plays in this structure: the d6 square is cleared for the king and the bishop looks at the knight on f3. It was a tactical error. The a6 pawn hangs, and the one move that still held the position was 31...axb5. After 31...Bc5 White stood a pawn and a half better and had a king attack to go with it.

The attack was 32.hxg5. After 32.hxg5 hxg5 33.Nxg5 fxg5 34.Bxg5+ Nf6 35.Bh4 White has two pawns and a completely exposed black king for the knight, and the count ends at a good pawn and a half. At this level about one player in nine finds the move; that is the number the human layer reports.

Gukesh played 32.Rab1, the rook move that roughly one in three plays. It prevents nothing bad and prepares nothing concrete. Objectively the game lost a good half pawn of practical clarity here; on the board there were still a pawn and a half to work with, but nothing that could be turned into a combination. When the opponent plays a developing move that is a tactical error, the answer is rarely the next developing move. It is the tactic the error now allows, and it has to be seen in the same calculation that spots the error.

Between move thirty-two and move thirty-four the game turned over twice. 33.Ne1?! came after nine minutes and sixteen seconds and melted the edge to a single move; b6 would have been that move. Muradli, also long in time trouble, replied with 33...Rhg8?, gave the inaccuracy back at once, and after 34.Nc2 put the rook on the wrong square with 34...Ra8?, where the a6 pawn would finally drop. On the board White had close to three pawns of advantage, on the clock one minute forty-seven.

The move that held it was 35.b6. After 35.b6 the passed pawn steps one square forward, backed by the pawn on a5, and the rooks can occupy the c-file at their leisure. On the calculation, 35.b6 Rg7 36.Rbc1 leaves White about a pawn and a half better. Objectively it was the move; practically it was the one move that anchors the count on a passed pawn rather than on a king step.

Gukesh played 35.Kf3?? and handed it all back. Objectively the move is a mistake of roughly three pawns; the word for it is mistake and not inaccuracy. The reply Black did not find was 35...axb5 36.Rxb5 b6 37.axb6 with a count around three quarters of a pawn for White, just enough to keep the position playable. Muradli played 35...Rgc8?? instead and put the advantage back to over two pawns for White; with one minute thirty-three on his own clock he did not calculate any differently than his opponent. In time trouble the move with a forcing point, here the push b5-b6, almost always beats the king that takes a half step. A king looks for a square, a passed pawn looks for the next square; only the second programme keeps running on its own.

After 39...Kc6 Gukesh had one minute fifty-seven left, after 40.Rh1 thirty-five seconds. Move forty bought him the thirty minutes of the added time control, showing on the clock one minute thirty-one after the ply; on the board the advantage had shrunk to a good pawn, and the position was no longer winnable but easy to hold. For six moves White manoeuvred without a concrete plan, and the edge drifted down to about half a pawn.

The one move that still carried the position was 47.hxg5 hxg5 48.Bd2, with roughly three quarters of a pawn left. Gukesh played 47.Kc2, and Muradli found the reply at once: 47...Ba3. The bishop blocks the white rook on b1 and covers the squares where the white king would look for shelter. 48.Rh3 changes nothing about this, and the sequence 48...Rb2+ 49.Kd1 Rb1+ 50.Ke2 Bc1 sets the perpetual. The bishop on c1 stops the white rook from stepping out with Rc2, the bishop on h4 has to keep an eye on the rook on g5, and the king shuttles between d1, e2 and f1. After 51.hxg5 Bxg5 52.Bh4 the position stood, and Muradli repeated three times.

Half a point in a match where the other top board offered Azerbaijan a full one; the count at the top asks for both together. Learn the patterns in which a rook and a bishop force a perpetual: the rook gives the checks, and the bishop takes away the one square the king would otherwise use to duck out. Without that one blocking square the pattern is not there, and with it the pattern cannot be calculated away.