18.f5 is a bit premature (inaccurate compared with simply 18.Rxc8) where if black played 18...Rxc1 19.Qxc1 first before 19...Nc6 black gets a small advantage.

After 22...Qf8 white's pawn on f6 looks impressive but the game is dead even, white's attack has no way to succeed without a blunder. 23...Qg5 was probably best but the retreat 23...Qh4 was also OK. 23...h5 was also probably best but also no big deal.

The sac 25.Rxc6 was objectively a mistake. Of the two moves, 25...Bxc6 punishes it best but after 25...Rxc6 black is still slightly better anyway. After 25.e5 another mistake black's advantage is now solid but he must play very accurately.

27...Rc5 was a counter-mistake giving back most of black's advantage. If instead 27...Rh8 followed by 28...Kg8 (after either 28.Qg3 or 28.Qg5 or Kg8 later if white makes yet another mistake), white's attack is abated with black up the exchange for insufficient compensation. For example, 27...Rh8 28.Qg5 Kg8 29.Qf4 and white will get back the pawn but have nothing more. Eventually, black will free the rook on h8.

click for larger view

Back to the game, after 28.Qg5 black's best was probably to simplify with 28...Bxg4 29.Qxg5 Rex5 holding an advantage. 28...Rh8 looks good at first (even if holding the tension was unnecessary) but here it was a bit inaccurate. The position is (unnecessarily) complicated here it helps to use an engine to make sure.

30...Bd7 was the best move for black but after 30...Ba6 the game is basically equal. 32...Rc1+ was a good plan but after 32.Kf2 black made a blunder with 32...d3 a try that was entirely too premature. Here, because of his earlier inaccuracy black should merely first nab the a-pawn (e.g., via 32...Rf1+ 33.Kg2 Ra1 34.Kg3 Rxa2) and then go for a draw by repetition with Ra1+ Kg2 Ra2+ etc. Note that black's queen can come in for a mate via a3 if white tries to avoid this.

Would 32...Rf1+ been enough for a draw? Beyond me, but surely a much better try than the hopeless push for promotion with 32...d3.