Black hasn't taken any advantage of the pressure on the f column, and this allows white to relocate his knight on e3. He couldn't if he played Bc4. Against Bc4, it seems playable d5, forcing white to retreat, even after the exchange. If exd5, cxd5 Bb3 (or a2) and black can play d4. Now, if white exchanges cxd4, the c column gets open to the rook's action... and if he doesn't, black can (d4xc3), leaving white with a very weak queenside. So I dare to say that 18. Bc4 is not a good idea - unless I miss something, of course...

Sounds reasonable. It is hard to see why White would want to play 18.Bc4, though...

Qh6 pins the white's knight, and prevents any ideas of g4 or Nh5. Black can follow with either b5 or Bb6, that forces white to defend. If white plays Kb1, black's queen can go to g6 and now, with the king on b1, b5 is even more effective.

Seems white should have better than 20. Rfg1 here. 20. g4 comes to mind.

Well, g4, Qh6, forcing white to defend the h3 pawn (h4, maybe?).

This time, GK engine seems to be right...

Too late for white to play Nf5+, but it is interesting. Black "has" to take BxN, since any other king's move allows, Ng5, hunting the queen (and now g5 is covered by the queen). But after BxN, white cannot cope with the threat Qa2 and retake the piece back. The now undefended c6 doesn't compensate white's weakness on b2, but it has to be seen what was coming out of a line like this: 23. Nf5+, BxN 24. exB, Qa2 25. Qc2, defending the bishop and b2, while white would go Bxc6 and Bb5, in order to contrast black's dominion on the b file (at a cost of a rook for black... still worth considering).

What's the point to drive back the rook? They will still keep looking at the b column. Again, Nf5+ and Ng5 (or Qg5) seem to be a little more promising, provided white cannot think for a central countergame, since d4 is impossible due to Nxe4.

Though this sacrifice isn't clear it does give plenty of room to work with. I'll will be interesting to see how you continue this line. White has some time to find a defense. I say best would be to leave a hole for whites king to mosey over to the e-f-files where it seems a bit safer.

rook for a bishop takes guts to pull off! I wouldn't do it here because I can't get my pieces into the attack that quickly... but then again, you;re a much better player than I am!  

Just wandering why not play Rxb3 instead and get an extra pawn thrown in with the exchange sac?

Now that you mention it; a very good question! Looking over it now, there seems to be no real objection to 25...Rxb3 at all (though I didn't really like 26.Bxb3 Rxb3 27.Na1 very much). What I had in mind was to leave a White a-pawn as a target at a4 for the d7-bishop after ...c5, which doesn't happen after 25...Rxb3. On the other hand, the a-pawn is immediately passed. The attack might have developed in ways similar to the game continuation; it might have been an altogether different game.

From the moment white hasn't provided room for his king to move (I think that Nc2 was the "big" mistake, but it is easy to say it now  , white is in real danger, both with Rxb3 or RxB. But the Na1 move seems to work pulling back the rook for the moment.

This is a move I'm sure most would miss, not sure I would have found it myself if I hadn't seen it.

White has a really nice looking counter sac here with Nxe5!

After 29.Nxe5 I had it in mind to play 29...Qf4+.

Brilliant sac! Note that the knight f3 "is not playing". The counter sac Nxe5 would work if after Qf4+ Ne3 black plays QxN, since at least white has Nd5+ and then NxB (exN is still there), if it can help (which I doubt). But if black plays dxN after the queen's check... nothing! since the knight is pinned.