16.) Qd3 looks weak to me and I can't understand what Nunn was thinking when he played it except maybe believing it'd stop c4. Obviously it didn't, as Shirov proved nicely.

18.) Ra1?, giving black a pawn for basically nothing it looks like.

...23.) Rxc4...interesting move by Shirov.

...27.) Kg7 I didn't understand why Shirov would have played this at first instead of the more knee-jerk Bg7 retreat. His 28th move illustrates.

16.) Qd3 looks weak to me and I can't understand what Nunn was thinking when he played it except maybe believing it'd stop c4. Obviously it didn't, as Shirov proved nicely...>

Overall Shirov played the whole game very strongly, nearly perfectly even, to punish White's wrong play. In his analysis in <Fire On Board> he maybe underestimated how good his position really was. He tried to prove that after White's best practical try 31. ♖xb3, 31... ♕xe2 was wrong and 31... ♕xb3 was right when probably both moves wins. The analysis of 31... ♕xe2 is very convincing despite the rather pessimistic final assessment, including the beautiful line 31. ♖xb3 ♕xe2 32. ♖f3! ♗g5? (♗e7!) 33. ♕c7 ♖f8 34. ♕xe5+ f6 35. ♖c7+ ♔h8 36. h4! ♕d1+ 37. ♔h2 ♕xf3!?

click for larger view

38. ♖xh7+! 39. ♕c7+ ♔g8 40. gxf3 which he gives as better for White (it's probably winning).

The analysis of 31. ♖xb3 ♕xb3 has some holes though. Shirov recommended 32. ♕xa6 a4 33. ♘g3 for White :

1°) However, the line 33. ♘g3 ♗e7 34. ♖c7 ♗d6 35. ♖b7 ♕c2 36. ♕b6 ♗c5 37. ♖xf7+ ♔xf7 38. ♕e6+ ♔f8 39. ♕f6+ ♔e8 40. ♕e6+ ♗e7 41. ♕g8+ ♔d7 42. ♕e6+ ♔c7 43. ♕xe7+ is obviously not a draw but a win for Black.

2°) In the same line, 35... ♕d1+ 36. ♘f1 a3 37. ♕b6 is still winning for Black with 37... ♕c2!! as there is no perpetual check after 38. ♕xd8 ♗c5 39. ♖xf7+ and Black ultimately wins the position with equal material thanks to the a pawn (this was quite impossible to figure out with the 90s engines).