A nine-move win, and the move you'd want back is the one where a winning knight jump was already on the board and you played a developing move instead.
Two of the three moves that mattered here were the engine's own first choice. The third cost about a pawn and handed your brother a defence he didn't take.
6.Nd4 — the move that made the game
After 5.d5 your knight on f3 was still hanging to the e4 pawn, and 5...Na5 gave you a free move to deal with it. Nd4 was the engine's top choice and it does four things at once: saves the knight, centralises it, hits f5, and leaves the a5 knight with exactly one useful square in the world. The continuation was Nxc4 Nxf5 Qf6 g4 Bb4 Qc2 at +1.5 — the c4 pawn goes, and the position doesn't care.
The general version: when a piece goes offside, the strongest reply is usually the one that takes away its route home rather than the one that attacks it.
8.e3 — the jump was already there
Forty-seven seconds, and +2.9 became +1.8. Ne6 was the move — Ne6 b5 Nxb5 dxe6 Bf4 exd5, +2.7 — and e3 is a sensible developing move in a position that had stopped being about development two moves earlier.
The cost was more than the number. After 8.e3 Black has Kf7, covering e6 and taking the fork off the board entirely: Kf7 b4 c5 bxa5 cxd4 exd4 leaves you much better but playing a game instead of finishing one.
At +2.9 the question isn't "what's a good move here". It's "what's forcing".
9.Ne6 — the same idea, one move later and twice as strong
8...Bg7 put a second piece on the fork square, and Ne6 hit queen and bishop with a pawn defending it. The reason Black can't just answer dxe6 is 7.Qa4: moving the d7 pawn opens the a4–e8 diagonal straight onto the king. So the engine's line was Bxc3+ bxc3 b5 cxb5 dxe6 at +3.0 — Black spending the bishop pair and his structure to buy the knight back — and David resigned instead.
Worth noting that the queen move that made this work was played two moves before the tactic existed.
Phase notes
The whole game was the opening. 5.d5 answered an attack on your knight by attacking something bigger, and the engine says Black only stays level by taking first with exf3. He didn't, and from move five you were never worse again.
Profile check
The captures-checks-threats scan is meant to run for your threats too, not just his — and 8.e3 is the clearest example yet of it not doing so. Forty-seven seconds in a position where a knight jump won on the spot, and what came back was a developing move. Pair it with the five-second h6 in the Caro-Kann earlier today and the pattern is consistent: when nothing is checking you, the scan doesn't start. The old habit of underrating your own position stayed quiet again — four games now.
What to drill next
- The position before 8.e3 — White to move, find Ne6 and play out Ne6 b5 Nxb5 dxe6 Bf4 exd5. Beat the engine from here three times running.
- Fork puzzles — knight jumps from your side of the board for once, rather than the ones you're avoiding.
- Trapped-piece puzzles — the 6.Nd4 idea: winning a piece by taking its squares rather than by hitting it.
When the position feels won, name every check and every knight jump you have before you let yourself play a useful move.