You won this game, but both of your own mistakes were trade decisions made on feel instead of on the line that follows.
Two moves in this game cost you real ground, and they are mirror images: on move 19 you traded a pair you should have kept, and on move 31 you kept a pair you should have traded. The move the engine wanted at 19 is the move you actually played at 31.
19.Re1 -- the rooks were doing your work
You brought the last rook to e1 and offered the trade. Twenty-four seconds, and the position went from +1.7 to +0.4. The game went 19...Qe8 20.Rxe4 Qxe4, and the engine's line reaches the same place: Rxe1+ Qxe1 Nb8 h3 a6 Qe4.
Qd3 was the move, and the point is not that it avoids a trade. It is what the rook then does: Qd3 Re8 Nf5 f6 Rd1 Qf7 holds at +1.6, with the rook stepping to d1 behind the d-pawn and the knight landing on f5. Your advantage here was space, and space is worth something only to the pieces that use it. Black's knight on a6 had no route into the game and his rook was his one piece with a job; taking rooks off removed the piece he most needed to trade and left the position with too little material for the d5 pawn to matter.
Carry this: with a space advantage, decide a trade by counting what stays on the board, not what comes off. If the trade helps the side with fewer squares, it is his trade, not yours.
31.Qd3 -- the same trade, three moves later and worse
30...Qe5 put the queen on a square where you could take it, and taking it won. Qxe5 dxe5 Ne4 Kg7 d6 cxd6 reaches +3.4: the recapture pulls the d6 pawn away, your knight comes to e4, and the d-pawn runs while Black's knight is still on a6, four moves from anywhere useful. That is the whole idea. The queens come off at the moment his worst piece is furthest from the action.
You spent fifty-two seconds and played 31.Qd3, and +3.0 became +0.0. What that allowed is the part worth sitting with: Nb8 Qe4 Nd7 f4 Qxe4 Nxe4. The queens come off anyway, three moves later, except now the knight has reached d7 and the position is level. You did not avoid the trade. You postponed it until it was his trade instead of yours.
The fifty-two seconds are the tell. That is not a clock problem, it is a question problem. The time went into finding a good square for the queen, when the question the position was asking was whether the capture in front of you wins.
Carry this: when you are winning and a capture or trade is available, calculate the capture first, all the way to a position you can name as won, before you look at any move that avoids it.
Phase notes
Opening and early middlegame. Eighteen moves went by with nothing flagged and the position went from +0.0 to +1.4. Part of that swing is begemon's doing, 10...Nb4 and 11...Na6 in particular, so the honest version is that you did not put a foot wrong for eighteen moves rather than that you built the advantage. Your openings have not been the problem for four games now.
Conversion. The advantage you gave back on move 31 came back on its own. 33...Kg7 and 34...Kf6 walked the king into the middle of the board with your queen and knight both active, and by move 35 the engine had mate in nine. Worth naming plainly: you won this on begemon's clock and begemon's king walk, not on the conversion.
Profile check
Same habit as last time, and this is the cleanest instance on file. In a winning position, once one candidate looks reasonable, the search stops. 31.Qd3 at +3.0 is the same shape as 18.c5 against berkut47 and 18.Rg1 against Russia-Melanesia, except that here you had fifty-two seconds and still produced a move that keeps the queens on rather than a move that wins.
What this game adds is the shape of the miss. It is specifically trades. Both errors were about whether to exchange a pair of pieces, and both times the decision was made on a general feeling, simplify while better on move 19 and keep queens on to attack on move 31, rather than on the position after the exchange. A general rule about trading is the enemy here; the line is the only thing that settles it.
What to drill next
- The position before 31.Qd3, White to move, begemon has just played 30...Qe5. Beat the engine from here three times running. You are looking for the capture, and then for the d-pawn.
- Convert a winning position. These are the positions where your search shuts down early, and the puzzle set is built out of exactly that moment.
- The position before 19.Re1. Same drill, three times running, but the target is different: find a move that keeps material on and improves a piece, and see how much easier the next ten moves become.
When you are winning and a trade is on the board, play the trade out in your head to a position you can name as won before you consider any move that avoids it.