You won this with the two hardest moves on the board — 18.Rd6 and 23.Bd5+ — but 22.Bxe4 shows the win was gone in between, and it was David's reply, not your technique, that handed it back. Out of a quiet Dutch you built a real pull, and then, near +3, went looking for a pawn.
18.Rd6 — the scan fired
David's 17...Qb6 was a mistake and you punished it at once: 18.Rd6, the engine's first choice, dropping the rook onto the sixth rank where it hits everything. The follow-up justifies it — Qd8 Rad1 Qe7 f3 b6 fxe4 pries the centre open and leaves White clearly on top. This is the winning-position scan working the way it should: David loosened a square, and your first question was what that move gave up. Nothing to fix here — this is the standard to hold.
22.Bxe4 — the pawn instead of the piece
The mistake, and an expensive one. Nearly three pawns up after 21...Nb6, the winning move wasn't a capture at all: b3, a cage. The knight on b6 has no safe square, and b3 Bd7 Rxb6 Qxb6 Rxd7 Qxb3 wins a piece outright, +3.3. Instead 22.Bxe4 took the e4 pawn and left the bishop hanging to the recapture; after fxe4 (the move David didn't find), b3 Bf5 Nxe4 Nd7 f3 leaves you down material with a loose king at -2.4. Thirty-seven seconds, and a won position became a lost one. The pawn was the useful move; b3 was the winning one — and this is twice now in two games that a winning position had a forcing idea on the board and you reached for the comfortable grab instead. Last week it was a knight fork left sitting at +2.9; here it was b3 trapping the knight at +3.1.
23.Bd5+ — the only move back
David returned the favour with 22...Nxc4, and here you were awake: 23.Bd5+ is the only move that holds, and you played it. David's 23...Kh8 walked into the finish — 24.Rxh6+ gxh6 25.Qxh6+ Qh7 26.Qxf8+, mating — and even the tougher Kh7 loses to Bxc4 Ra7 Bd5, White +5.7. The radar that missed b3 one move earlier was fully switched on the moment the position turned forcing again.
Phase notes
The opening pull slipped once: 6.Bg5 let David equalize (+0.7 to level), where Nf3 first would have kept the bishop's options and the edge. Small, but it's the difference between pressing from move six and pressing from move seventeen.
Profile check
Same question as last game: when you're winning, do you look for the move that wins or the move that's useful? Last week a knight fork went missing in a winning position; this week b3 did. It's a real pattern now, and it has a precise shape — it happens in quiet positions that are already won, with a free pawn on offer. That's the one thing this game says to fix.
What to drill next
- This position — White to move, +3.1. Find b3 and win the knight on b6; beat the engine from here three times running before you move on.
- Convert a winning position — puzzles that reward the forcing finish over the tidy one, which is exactly where 22.Bxe4 went wrong.
When the position feels won, name the move that wins — the check, the trap, the capture that ends it — before you reach for the free pawn.
Two honest questions — and in both, the move you talked yourself out of was the better one, for the same underlying reason. I ran the engine on the lines you were weighing; here's what it says.
The bishop you traded on e4
Your read was right: that bishop was locked out. On g2 it looks down the long diagonal and runs straight into the pawn on e4, with only f1 and h3 left to it. Wanting to fix your worst-placed piece is good thinking — the trouble was the method, in two layers.
"Two pawns" was really one. After 22.Bxe4 the bishop does hit e4, f5 and b7 — but David just plays fxe4 and takes it, and you've handed over a bishop for a single pawn. The second pawn only exists if he declines the recapture, and he won't. That's the whole -2.4.
And you didn't need to solve the bishop at all. Near +3, b3 holds everything together: it defends the c4 pawn David's knight was hitting, keeps your rook dominating on d6, and stays +3.3 (b3 Bd7 Rxb6 Qxb6 Rxd7 Qxb3). When you're winning by that much, a passive bishop is a luxury problem, not an emergency. And it wasn't even a bad piece — it was a monster waiting for the position to open, raking b7 and the a8-rook the instant the e4 pawn goes. It was one break (f3) from being your best piece. You don't sell a piece like that for a pawn; you open the door for it. That's the +3.1 position if you want to sit with it.
The 6.Bg5 question — Nf3, and the pawn you feared
This one is barely a mistake — the smallest flag in the game, +0.7 down to level — but the reasoning behind it is worth unpacking.
On castling: the dark-squared bishop was never what blocked O-O — the knight on g1 was. Nf3 develops the blocker and lets you castle next move, so it actually served your castling goal better than 6.Bg5 did. It's also the engine's top move here, +0.7.
On dxe5: the outpost fear was overpriced. You played 7.dxe5 yourself the very next move and were completely fine, and David recaptured with the d-pawn, not a knight. Even after a knight recapture, a knight on e5 isn't permanent — it gets challenged with tempo.
On Nf3 and ...e4 — this is the one that matters. You rejected the engine's best move because of ...e4, but ...e4 doesn't trap the knight, it overextends. After Nf3 e4 Ng5 (or Nh4) the knight simply reroutes and f3 undermines the pawn — the engine has White at +0.8 there, better than before the push. And you proved it yourself: when ...e4 finally arrived in the game at 15...e4, you rerouted with 16.Nd4 and were fine. The threat that steered you off the best move was worth less than it looked.
The habit under both
In one you let a piece's problem, in the other a scary-looking pawn, push you toward the softer move. It's the same fix, and it's the cheapest rating points you'll find: price the threat before you believe it, and when you're already winning, play the move that wins over the move that tidies up.