Game #514: my London System vs g5, and a 148-point upset
On June 6, 2024, in game #514 of my rapid journey, I beat Avazbek1999 (1093) with White and went from 933 to 945. The gap was 148 points, the ninth-biggest upset in my first 2,116 games. The opening was my London System, and Black met it with an early ...h6 and then the pawn push every London player runs into sooner or later: London System g5, the pawn storm that chases the bishop on f4. That part I handled well. What came after is the reason this page exists: I won material three times and gave it back three times, and the game ended on Black's clock, not on the board.
Where I was that week
This was week 22 of the journey, the week of 91 blunders. Two days before this game I had started a run of seven losses in a row, games #504 to #510, which took me from 974 down to 923. I broke it with two wins against the same player, Ras-stafari, and then the morning of June 6 I lost game #513 to 90stx from a position the engine rated +8.71. So I sat down for #514 with one lost win fresh in my head, against the highest-rated opponent of my week.
| Game | #514, June 6, 2024 |
| Opponent | Avazbek1999 (1093) |
| My colour, rating after | White, 945 (from 933) |
| Time control | 15 minutes + 10 seconds per move |
| Opening | London System (D02) |
| Result | Win on time after move 33 |
| Accuracy (Stockfish) | Me 73.5, Black 75.8 |
| My blunders / mistakes / inaccuracies | 4 / 3 / 2 |
Yes, my opponent had the higher accuracy. I won anyway, and I'll come back to why.
The opening: London System vs g5
The first moves were my usual setup: pawn to d4, bishop out to f4, knight to f3, pawns to e3 and c3. Black copied part of it with a bishop on f5 and a knight on c6, and then pushed the h-pawn one square and the g-pawn two. That g5 push attacks the bishop on f4 directly, and it is the main idea Black has against the London when they want a fight.
Stockfish preferred the retreat to g3. My bishop to e5 was its second choice and was not marked as a mistake: the evaluation stayed about equal (+0.44). Black answered with the bishop to g7 to challenge mine on the long diagonal. That diagonal decided the game a few moves later.
I then played the queen to a4 and the bishop to b5, pinning the knight on c6 to Black's queen. The engine agreed with every one of my moves from move 9 to move 13: knight to b3, bishop takes on c6, knight to c5, queen takes the pawn on c6. Black's queen had gone to d7 and then back to d8, and my opening accuracy was 93.3. For once, the London did exactly what I wanted from it: simple moves, a pawn up, no danger.
Moment 1: the knight that left d5 alone
On move 13 Black moved the knight from f6 to d7. It attacked my knight on c5 and opened Black's own bishop on g7 against my bishop on e5. The engine called it a blunder: +1.31 to +5.53. The problem was that the knight on f6 had been the only defender of Black's pawn on d5.
This one I found. Bishop takes the bishop on g7, the king takes back, and my queen takes the pawn on d5, attacking the bishop on f5 too. It was the engine's line move for move, and after 15.Qxd5 I was +5.34, two pawns up with Black's king out in the open on g7. That's what the ...g5 push had cost Black: once the dark-squared bishops were gone, the squares around the king had nobody to cover them.
Moment 2: the pawn on g5 that wasn't free
Black put the bishop on g4, attacking my knight on f3. I looked at the g5 pawn, the one that had started the whole fight, and took it.
The pawn on g5 was guarded by the pawn on h6. After my knight took it, Black could simply take the knight back, and the evaluation dropped from +5.31 to +2.20. The engine's move was knight takes knight on d7, followed by a queen trade, and it still rated that +5.17. The trade is the boring move. I keep writing that sentence in these posts.
Moment 3: the check I liked better than the capture
Black didn't take my knight. On move 16 Black pushed the pawn to e6, attacking my queen, and the engine called that a blunder too: +7.18 for me, my best moment of the game.
The knight on d7 had no real protection: if Black's queen took my queen there, my knight on c5 would take back. Queen takes on d7 kept everything (+7.24). I saw a knight check that attacked two things at once and played it. The bishop on g4 simply took the knight, and the evaluation fell to +2.72. I then took on e6 with the other knight, with check, instead of the queen (another mistake, +1.32). After the smoke cleared I had queen, two rooks and eight pawns against queen, two rooks, a knight and three pawns. Better, but not the crushing position I'd had two moves earlier.
Moment 4: the pawn push instead of the queen trade
Black blundered again on move 22 with the queen to d5, and I took the c7 pawn with check (+4.31). Then I drifted: a pawn to c4 (inaccuracy), a pawn to g3 (mistake), and on move 26 this.
The engine wanted the queen trade. I pushed a pawn at Black's queen instead, and +3.15 became +0.16. For the next five moves the position stayed close to level, sometimes slightly better for Black, until Black's pawn took on g3 on move 31 and I was a little better again (+1.58).
Moment 5: the rook move that won a knight, and the move I played
On move 32 Black's queen took my pawn on f5. It was the last move Black made, and the engine called it a blunder: +7.00.
Rook to f1 attacks the queen, which had no defender, with the knight standing behind it on the same file. After the queen moves, the rook takes the knight. The engine rated it +7.52. I pushed the c-pawn, and the evaluation went to -0.30.
Then nothing happened. Black's clock showed 12:39 after that last queen move, and Black never made a 33rd move. The game record says I won on time. I don't know why my opponent stopped playing, and I won't guess. What I know is that the engine had the final position about equal.
What I take from game #514
The London System vs g5 part went well, and I'd play it the same way: don't panic when the pawn hits the bishop, keep the bishop active (e5 was fine, g3 was the engine's pick), and if Black's dark-squared bishop comes to g7, a trade there leaves the king with holes around it.
The rest is the same lesson as my worst losses, only this time it didn't cost me. Four times the engine's move was a plain capture or a queen trade: knight takes on d7, queen takes on d7, queen to f4, rook to f1. Four times I played the move with more noise, a pawn grab, a knight check, a pawn push. With a big advantage, the quiet capture is almost always the right one.
And the honest part: a 148-point upset with 73.5 accuracy and a level final position is not proof that I'd suddenly become a 1093. It's proof that rapid games at my level are decided by who keeps going.
What came next
That evening I lost game #515 to oceano929 (1087), my first loss against the Grob, after beating it as Black in game #396. Then I won two in a row and finished the week at 956. The next week, week 23, I got back over 1000. For another London game where the attack actually worked, see game #216; the whole series lives on the journey page.
If you want your own games read like this, with the capture you passed on and the reason in plain words, that's what Chess Coach does inside Claude: here's how the analysis works. You can replay game #514 on Chess.com.
Frequently asked questions
What should White do in the London System when Black plays ...g5?
In my game Black played ...h6 and ...g5 to chase my bishop on f4. Stockfish's first choice was retreating the bishop to g3; I played bishop to e5, its second choice, which kept the position about equal and led to a trade of dark-squared bishops on g7.
How did a 945 beat a 1093 in game #514?
Black left the pawn on d5 without its knight defender on move 13, and the bishop trade on g7 won it. I was +5.53 after that, but I gave most of it back with flashy moves. The game ended on time: Black still had more than twelve minutes and never made a 33rd move.
Was game #514 a well-played win?
No. Stockfish gave me 73.5 accuracy and my opponent 75.8, with four blunders each. When the game ended, the engine rated the position about equal, at -0.30.