Not so big mistake

The only reason to play Bc3 is to avoid ...Nxb2 (...NxB). But that exchange is not so undesirable for White, as it would remove an active knight. Black meanwhile, faced with the threat of Bxc4 dxc4; Qxc4 ... has to decide between the exchange or retreat.

seems reasonable

@Joe - Don't sell the computer analysis too short. It has its limitations, yes, but for short-range tactics it will be reliable enough. After 20...g4 the f3-knight's only retreat is 21.Nh4.

both moves seem sound!

Weakening White's castle. Better was Qxf3 with a winning position

I generally have a rule about using pawns in front of my king. Taken on a case by case basis obviously, but you usually don't move those and expose your king.

In a way the computer DID report the mistake: check out evaluations before and after this move: +3.50 gets reduced to +2.83. This capture is not only weakening (not so much a consideration given the inherent strength of White's position) but also interferes with the mobility of White's pieces. But it is not all bad: that half-opened g-file might yet come in handy.

Qxf3 was best, have to agree best not to weaken the castle position.

Yes!

Agree! 24.Ra6 Rc8 25.Rc1. That pawn on c6 is a serious Black weakness and a big, big target.

looks good to me.

The final position of your analysis is fantastic! Black has enough compensation for the exchange

The threat of ...Bxe5 goes further than the win of a pawn, but targets the b2-bishop, which is pinned against the empty b1-square. So Bxe5 Rb1#. How would White save her bishop then? To answer the threat of ...Bxe5, White might have to play something like 29.h3 or h4. Even then Black would have had much of the play, e.g. after ...Nd2. Nice line.

There is an interesting possibility here - 27.....Qb1+ 28. Rxb1 Rxb1+ 29. Kg2 Nf4+ 30.Kg3 and Rg1 mate - unfortunately white can improve with 29. Bc1 with an easily won game for white. gameknot.com

I like what was played!

I think I prefer 27.Ba3 here. Maybe the rook disconnect is a problem?

So the rook disconnect is - or was - a problem after all. Black has a surprisingly active game with plenty of counter-play.

The longer mate nets are tricker to find, all I can recommend is practice unfortunately

Really, all you have to see is that there is a forcing line, and then calculate along it. An experienced tactician would at once check out 31.Rxg7ch to see whether it promised anything. Given the open 7th and 8th ranks, and the bishop check as well - looks very promising - then White would look further along. Noting that after 32.Bf8ch, ...Kg8 is the only one that doesn't lead to a forced mate in two, the response 33.Bb4 attacking King and Queen at once would have sold me. So I would have played the line without having seen the 6-move mate! Chances are that I would have seen it in time NOT to play 34.Bxd2, though! Given the presence of a forcing line, with practice you will be able to 'see' tactical ideas and 'calculate' consequences with little trouble. It does require practice though!

I need practice on this, but I do like the R sac that you mention, and Whites DSB is in a good location, and Blacks DSB was pinned by the R.

Fortunately, Black's attack isn't quite enough!

just goes to show how valuable the N is.

There was nothing doing with 32.Rxg7ch? Too slow: E.g. 32.Rxg7ch Nxg7 33.Bf8 Rb7 34.Bxg7 Rxg7 35.Qxe6 ... It might have led to an advantage in the major piece end game, but I think I'd prefer, as White has, to keep the material edge and perhaps trade it later for a better return.

this move removes the pin on the B

37.Bb2!! is a KO punch

37.Bb2! ... [A] 37...Qxb2 38.Qc8ch Kh7 39.Ra7ch Ng7 40.Qf5ch Kg8 (or...Kh8; Qf8ch) 41.Qf7ch and mate next. or [B] 37....Rxb2 38.Ra8ch Kg7 39.Qf8ch Kg6 40.Qe8ch Kf5 41.Qf7ch Kxe5 42.Re8ch Kd6 43.Rd8ch Ke5 (else ...Kc5{relatively best} 44.Qa7ch Rb6 {Kb5 Rxd4; Rb1ch Kg2; Nf4ch Rxf4}45.Rxd4 and White wins) 44.Qe7ch Kf5 45.Rf8ch! Nf6 46.Rxf6ch Qxf6 47.Qe4#;or [C] 37...Qf4 38.Ra8ch Kg7 39.Ra7ch Kg8 40.Qh7ch Kf8 41.Qh8#. ******************************************* Line [B] turned out tougher to work out than I thought at first.

38...Kh8? is a bad mistake. After 38...Ng7 Black can hold

After 38...Ng7, Black might even have had slightly the better of it. This a real knife-edged game!

After 40.Rad1, ...Rb8 41.Qxh5 Qxe5 42.Qxh6ch. White's two extra pawns in the major-piece endgame don't look so conveniently placed.

43.Qg6 c5 44.e6!? ... (not sure about the value of this move, but it's worth looking where it might lead) 44...c4 45.Qe4 c3 46.Qd4ch Qg7 47.Qd8ch and now [A] 47...Qg8 48.e7 c2 49.Qxg8ch Kxg8 50.e8=Qch ... 51.Qc6; or [B] 47...Kh7 48.e7 c2 49.Qd3ch Qg6 50.Qxg6ch Kxg6 51.e8=Qch etc. ******* Queen endings are tricky things, and the line I chose for Black in this 'sample' line may be far from Black's best option(s). But I think it suggests the outline of a plan for White, and some of the motifs and patterns indicated here are interesting and useful.