Abdel Irada: <<•>Do or die?<•>>White has a deadly threat: mate on h7. And taking the g-pawn doesn't help; then it's mate in two. So in this case at least my old junior high football coach is right: "The best defense is a good offense."
Fortunately, Black does have the move, and after the initial check, it's knighty-night for White. (Or at least so it *appears*; but appearances can deceive.)
<<•> 27. ...Nc3† >
White has two moves worth examining.
Out of the question are (a) 28. Qxc3?, Rxc3 
and (b) 28. Ka1??, Ra4#.
< (1) 28. Kc2, Rxb2†!
29. Kxb2 ... >
White loses decisive material after either (a) 29. Kd3?, Rxh2 30. Qxh2, Nxd1 or (b) 29. Kc1? and the same sequence.
< 29. ...Qb4†
30. Kc2 ... >
Not playable are (a) 30. Ka1?, Qa3† 31. Ra2, Qxa2# and (b) 30. Kc1??, Ne2#.
The text, however, is no better.
< 30. ...Ne2††
31. Kd3, Qe4# >
And so much for line (1).
<<•> (2) 28. Kc1, Ne2†† >
Black can go fatally astray here by getting greedy: (a) 28. ...Nxd1† 29. Kxd1, Rd4† 30. Bd3, Rxd3† 31. Qxd3, Kxg7 32. Qxh7†, Kf8 33. Qh8#.
Now White must make another choice.
< (2.1) 29. Kd2?, Rxb2†
30. Ke1 ... >
Not (a) 30. Ke3/Kd3?, Rc3†, winning the queen.
< 30. ...Qb4†
31. Kf2, Nxf4† >
Again Black wins the queen.
*However*, White may be able to draw.
<<•> (2.2) 29. Kb1 ... >
Here I thought Black could break through with 29. ...Rxb2†, but I find no follow-up after 30. Kxb2 and either (a) 30. ...Qb4† 31. Qb3 or (b) 30. ...Rb8† 31. Ka1 and White escapes since his queen controls a3.
If this holds up, Black will have to settle for 29. ...Nc3†, with a draw by repetition.
On the other hand, since I didn't sleep (again) last night (Ah, the joys of intractable insomnia!), it's quite possible that I'm suffering amaurosis schacchisti. Wouldn't be the first time. :-S
(Note: I've also considered 28. ...Na4† 29. Kb1, Rxb2† 30. Rxb2, Nc3† 31. Kc2, Qa3 32. Bd3, but there appears to be too much sand in the gears of my brain. There may be something here, but I'm not finding it right now.)
sac: 29...Rxb2+
.
