Four moves cost you something in this game and every one of them was a queen move; in two of those the engine wanted the exact move you played, one turn later.
You won this comfortably in the end, and the last ten moves were the best chess on the board. But between move 15 and move 17 the evaluation went +1.8, +0.3, +2.3, +1.0, +0.0, and only one of those swings was Gugabud's doing.
15.Qc2 — the rook had to go first
Twenty-seven seconds, and +1.8 became +0.3. The queen was not on a bad square. She was early. The engine's line is Rc1 Kb8 Qc2 h4 Nxc7 Nb4, holding +1.8, and your move sits inside it two turns later. What the rook does on c1 is make Nxc7 possible: Gugabud had castled long, and after Kb8 a knight landing on c7 stands next to his king and cannot be taken, because the rook behind it covers the square. Play Qc2 first and the knight has no such landing.
What you allowed instead was Nxd4 Nxb6+ Qxb6 Bxb7+ Kxb7 exd4, a full liquidation that burns off most of what you had built. He did not find it, which is luck rather than a defence.
Carry this: when a quiet move and an active queen move both look right, check whether the quiet one creates a threat the queen move is going to need. If it does, it usually goes first.
16.Qf5+ — a check that improved his king
Forty-three seconds spent on a check, and +2.3 went to +1.0. Kb8 was his only reply and it was also the move he wanted, because it stepped his king off the c-file, which is the file your rooks then spent fifteen moves fighting to use.
The move was Nxb6+ cxb6 d5 Kc7 Rac1 h3, which the engine puts at +2.7. Nxb6+ is a genuine sacrifice and its point is the recapture: cxb6 pulls the c7 pawn away, so the king is standing on an open file by the time Rac1 arrives. The d5 push in between shuts your own bishop's diagonal but hits the knight on c6 first, which is what buys the time.
Your check did the reverse job. It relocated his king to a safer square at no cost to him. Before inserting a check, run the line you actually want with the checking move deleted, then ask whether the square his king lands on is worse for him than the one he was on. Here it was better.
17.Qxg5 — the right capture, one move early
Twenty seconds, +1.2 to +0.0, and it is move 15 again in a purer form. The engine's line is Nf6 Ne7 Qxg5 Bxg2 Kxg2 Qe6, worth +1.2, and your move is the third move of it. The pawn on g5 was not running anywhere. Nf6 forces Ne7 first, and only then does the queen take, with his knight sitting on a square it did not choose.
Then Gugabud played 17...hxg3 in four seconds, handing back +0.0 to +3.1, and you found the only move that holds it: 18.fxg3. That deserves saying plainly. There were two recaptures and the tidier one throws the game away; fxg3 opens the f-file and the engine's continuation uses it immediately with Na5 Rf5 Qc6 Qf4 Rd6 h4. You took the structural damage and got a file for it, which is the same judgement this review keeps saying you skipped two moves earlier.
Carry this: before you take a pawn that cannot run, look for the move that forces a reply first.
31.Qxd8 — the trade that gave back a mate
Gugabud played 30...Kb6 and the position stopped being merely winning. It was mate in five: R1c6+ Ka5 Qf5+ Re5 Qxe5+ Qd5, and the first move is the whole idea, because his king had just stepped onto the sixth rank where a rook arriving on c6 gives check.
You spent forty-one seconds and played 31.Qxd8, trading into a rook endgame at +5.5. That is still winning by a distance and you did win it. But the forty-one seconds went into deciding whether to trade queens rather than into asking what his king move had just allowed, and the answer to the second question was on the board in one move.
Carry this: when their king moves, look for your checks before you look at anything else. A king move changes which of your pieces are already pointed at it, and that is cheaper to check than any plan is to evaluate.
Phase notes
Opening. 7.Nge2 developed in a position where d5 was free for one move. After 6...Bd6 his own bishop blocked the queen's defence of the pawn, and Nxd5 Bb7 Bxd6 Qxd6 Nf4 Qb4+ takes it and holds +1.7. Twenty-four seconds, and it slipped to +0.4.
Endgame. From move 32 the engine flags nothing at all. 33.R1c3 and 35.Rd3 put a rook behind the passed pawn, 37.d5 and 38.e4 rolled the centre, and 39.Rc7+ Kd6 40.Rc6+ Ke5 drove his king onto the square 41.Re6# needed. This is the phase you have least to worry about.
Profile check
Is move 31 the same mistake as last time? Yes, and closely enough to be uncomfortable. Against begemon you spent fifty-two seconds on 31.Qd3 at +3.0 and declined a capture because keeping queens on felt right. Here you spent forty-one seconds on 31.Qxd8 and took a trade because taking queens off felt right. Same move number, opposite instinct, identical failure: the exchange got settled by a general rule instead of by a line you calculated to a position you could name. Which way the rule points is not the problem.
One to add. Moves 15 and 17 are the first clean cases on file where the engine's improvement contained your exact move, one turn later. That is the in-between-move habit facing outward, and 16.Qf5+ is the same habit facing inward, three moves earlier, the way 14...Ne2+ against chessydonmf did. Both directions in one game is new, and it points at a single fix rather than two.
What to drill next
- The position before 31.Qxd8. His king has just gone to b6 and there is a mate in five. Find it, then beat the engine from there three times running. This is the one for this week, because it is the repeat.
- Zwischenzug puzzles. Moves 16 and 17 are both this theme in opposite directions, and it is the cheapest item here to practise in volume.
- The position before 16.Qf5+. A knight sacrifice on b6 against a king that castled long. The exercise is to find something better than a check before you play the check.
Before the queen moves, name the quiet move that would make her arrival stronger a turn later, then play that one first.