Game #265: how to play against the Levitsky Attack, from my win against a 1156

On March 5, 2024, in game #265 of my rapid journey, I beat danielnice (1156) with Black, rated 994 before the game. White opened with the Levitsky Attack (1.d4 d5 2.Bg5), and how I played against it was simple: I asked the bishop a question with h6, developed everything, and waited. The position stayed about equal for 22 moves; then White made one blunder and I found the move that punished it. My rating went to 1006, the highest it had ever been, and the 150-point gap makes it one of the ten biggest upsets in my first 2,116 games.

It's also the cleanest game in this series so far. Stockfish found no inaccuracies, no mistakes and no blunders in my moves, for an accuracy of 95.9 (Chess.com's own review gave 91.85). I'm writing it up because I want to remember what a clean game felt like from the inside: quiet, a bit boring, and then over in eleven minutes.

Where I was that week

This was the second day of week 9 of the journey, the first week I finished above 1000. I'd touched 1001 once, a week earlier, and lost it the very next game (the story of that is here). Since then I'd been stuck in the 970s to 990s. Two games before this one I drew with kabaha01 (1153) and sat at exactly 1000; the game right before this one was a loss to VasekExce-inactive (1078), back to 994.

The Levitsky Attack: asking the bishop a question

The Levitsky Attack is what Chess.com calls the start 1.d4 d5 2.Bg5: White's bishop comes out early to g5, before any knights. My answer was to push the h-pawn and kick it.

r n b q k b n r
p p p . p p p p
. . . . . . . .
. . . p . . B .
. . . P . . . .
. . . . . . . .
P P P . P P P P
R N . Q K B N R

Stockfish's first choice was the pawn to c5, attacking White's center. It rated both moves about equal, -0.12 for c5 and -0.09 for my h6 (evaluations are from White's side, so a minus is good for me). The bishop retreated to h4, and I developed my other bishop to f5, outside my pawn chain. Again the engine slightly preferred c5, and again it called the position about equal. My accuracy in the opening was 96.7.

The trades came on moves 8 and 9. White's knight took mine on d7, and White's bishop took my knight on f6. I recaptured with the g-pawn, which doubled my pawns on the f-file but opened the g-file in front of my rook. Later in the game my rook went to g8 and then took a pawn on g4 down that file. So the bishop that came out on move 2 ended up traded for a knight, and both of us castled on the queenside.

The quiet middle, and the move I didn't play

From move 12 to move 22 nothing dramatic happened. Both of us pushed pawns on the queenside, in front of our own kings. The evaluation stayed between about 0 and -0.9 the whole time. Then, on move 22, I had a choice.

. k b r . . r .
. . b . q p . .
. p p . . p . p
. . . p p . . .
p P . P . . P .
P . . B P P . .
. K P Q . . . P
. . N R . . . R

Stockfish wanted the e-pawn forward to e4, a double attack on White's bishop and the f3 pawn. I played another queenside pawn move instead. It wasn't marked as an inaccuracy (-0.63 before, -0.22 after), but the engine's idea was sitting right there. One move later it came back, much stronger.

Moment 2: the knight that trapped its own bishop

On move 23 White brought the knight from c1 to e2. Stockfish called it a blunder: the evaluation went from -0.22 to -5.03. The engine's move for White was the bishop to h7, attacking my rook on g8.

. k b r . . r .
. . b . q p . .
. . p . . p . p
. p . p p . . .
p P . P . . P .
P . . B P P . .
. K P Q N . . P
. . . R . . . R

This time I pushed the pawn. Look at the bishop on d3 after that: its own pawn is on c2, its own knight now sits on e2 and blocks the way back, my pawn on e4 is attacking it, and the square c4 is covered by two of my pawns. Its other moves were captures on b5 or e4. In the engine's line White gives the bishop for a pawn on b5, which is exactly what happened in the game: White took my pawn on e4 with the f-pawn, I took back with the d-pawn, and White's bishop took my pawn on b5 and was captured by my c-pawn. I was a piece up for a pawn.

I can't tell from the record how much of that I saw. What the move clearly was, even at a glance, is a pawn attacking a bishop and a pawn at the same time, and that alone made it worth checking. That's the useful part: a double attack with a pawn is worth looking for every move, because a piece can lose its escape squares with a single quiet move like White's knight to e2.

Finishing it: the open g-file

After that the game went fast. White's knight came to f4; I took it with my bishop, and Stockfish rated my capture almost exactly as good as its own preference, taking the g4 pawn with the bishop. Then my rook took on g4, down that g-file the trade on f6 had opened, and later took White's f4 pawn too. My e-pawn kept walking: e3, then e2.

White's queen check on b5 didn't change anything. On move 34 White's queen took my pawn on e2, and my rook took White's rook on e1 with check, attacking the queen at the same time. White resigned. The game took about eleven minutes; I still had 13:29 on my clock. You can replay the full game on Chess.com.

How I'd play against the Levitsky Attack now

From this one game, not from theory:

• Ask the bishop a question early. The pawn to h6 was fine by the engine's count, and so was c5. What mattered was that the bishop either had to retreat or trade itself.
• Don't fear doubled pawns after the trade on f6. Taking back with the g-pawn gave my rook an open file it used twice.
• Develop and wait. I didn't need a plan against the opening. The game was about equal for 22 moves, and the win came from one blunder by White and one double attack by me.

My middlegame accuracy was 95.6. Compared with my other upsets, the difference wasn't a brilliant idea; it was that I didn't give anything away.

What came next

danielnice and I played twice more straight after. I won both: game #266 with White, and game #267 with Black against the Levitsky Attack again. Three wins in a row took me to 1028. The next game was a loss, but the week ended at 1023, the first time I finished a week above 1000. The whole week is in week 9 of the diary, and the full series is on the journey page.

If you want your own games reviewed like this, Chess Coach runs Stockfish on them and explains the key moments inside your AI assistant: here's how the analysis works, and what Claude can and can't do with chess games.