I found the same line as <al wazir>, by chance as the first line that I looked at.

19. Ng6+ demands to be considered. It exploits the pinned g pawn, shoves an attacking knight up the board (most great combinations rely on a far advanced knight) and it owns two of the squares around the black king (f8 and e7).

19. ... Ke8 is forced. Now we have a stalemated black king, and the rule for stalemated kings is to check, check, check until dead. The check has to come on f7, so let's try 20. Qxf7+ Nxf7 (forced). Now we need another check on f7, so let's look at Bg8. The knight cannot be defended, so we are home free.

No, wait, the knight can dodge out of the way and still defend f7 from its new post. Ah, that's okay we have 21. Rf8 mate.

You have got to love the plight of the black bishop on d7. Black would dearly love to move it, either to defend f7 or to allow the queen to defend along the seventh rank. But it is totally hemmed in by its own pieces on all four compass points.

Initially, my idea was then 21.Bg8 (threat:Bxf7+ with mate to follow), but I think black had one defense: 21...Nf5 (blocking the Q+R battery and putting my knight en prise).

But I had so much firepower attacking the kingside that I figured a queen-sac was worth looking at, so: 21.Qxf7 Nxf7.

1. First the ♕ sac: 20. ♕xf7 ♘xf7 21. ♘g6+ (and not ♖xf7 as proposed by <dezechiel>) ♔e8 22. ♖xf7 ♔xf7 23. ♖f1+ ♔e8 24. ♖f8#

First the ♘ sac (see game)

Moving the N on f7 leads to Rf8#. If the N is left on f7, then it will be captured by the white B, which is also mate. I can't see any way for black to prevent this capture. The black bishop is hemmed in and can't move, the black queen is blocked by the bishop, and the rooks can't do anything.

Now, going back to review that idea, I see that it has some merit. Obviously 21...Nxe4 is out due to 22.Qxf7#, but for the same reason, black can't allow 22.Nxd6+ either.

That leaves black with blocking the f-file with <21...Nf5>. But now white can attack the knight with <22.g4!>. This time, black cannot take the white knight as 22...fxg6 23.Bxg6+ Kf8 (not 23...Ke7 24.Qc5#) 24.gxf5 wins a knight, and mate will be served up soon.

The idea of playing Bg8!! is beautiful in that it mates a move faster and that making quiet moves in the middle of a combo like this always seem a bit more esthetically pleasing to me.

For those who wanted to play Ng6+ Ke8 Bg8?? you lose. You need to sac the Queen first so the knight can't go to f5. The suggested Qc5 simply loses to fxg6!

20. Nh6+ Ke8 21. Nh8 Nf5 and White no long has access to the f7 square. The attack is blocked, and black is left in a better position than before.

21. Nh8 also loses to 21. ...Ke7 (although thats avoidable) and 21. ...f6

Here's my thinking process:

Uhm, 20. Ng6+ looks a natural try, but after 20... Ke8 I don't think there is much white can do.. The Nd6 is the only defender of f7, so maybe I can deflect him, so let's try 20. Ne4. Now if 20.. Be8 to protect f7, then 21. Ng6#. 20..f6 or 20..f5 are obviously suicidal (21. exf6 ), so 20...Nf5 is forced, but after 21. Bxf5 exf5 22. Nxf5 white is clearly better (a pawn up, a very strong attack, several threats and better placed pieces)...

After checking the game:

Ng6+ Ke8
Qxf7+! Nxf7
Rxf7 Kxf7
Rf1+ Ke8
Rf8#

Material even. Black's K has no defense. Black's heavy pieces all have access down the half-open (soon to be fully open) f-file. White has two ways to directly attack the Black K 20.Ng6+ and 20.Qxf7+. Both are absolutely forcing.

20.Qxf7+ Nxf7 21.Ng6+ Ke8 22.Rxf7 Kxf7 (Black can't protect Rf8#) 23.Rf1+ Ke8 24.Rf8# 1-0

Pretty obvious from the cramped K position, and heavy piece access to the enemy K. Just for fun, let's see if the N check works too.

20.Ng6+ Ke8 21.Qxf7+ (continues as before) Nxf7 22.Rxf7 Kxf7 23.Rf1+ Ke8 24.Rf8#

Most of us will readily confess--we only get these puzzles because we're put at a starting point and told to look for a solution. If we reached the actual position(s) in one of our own games, we would quite likely not see the themes that seem to emerge from these selected puzzles.

I try to approach them as if they were part of a game. I ask myself - would I have spotted that a combination was "on" and so invested the time to calculate more deeply? And what are the features in the position that make a combination possible - stalemated king, weak pawns, unprotected pieces etc? If I can spot those features, I am more likely to be able to create them in my own games.

The move that Black really seems to have overlooked was 17. Bxh7+!. This sacrifice always needs to be considered when White advances a pawn to e5, thereby driving the Black Knight from f6 and opening the Bishop's diagonal to the h7-square, especially when White has a Knight on f3 (ready to jump to g5) and his Queen is able to move immediately to the h-file. If Black had played 16. ... h6 (instead of 16. ... Bxg5?), his position would have been only slightly worse.

The position after 16. ... Bxg5 is a good exercise for analysis. If given as a puzzle, one would almost assume that the sacrifice 17. Bxh7+ "must work", but it is not so easy to verify. Here are the critical lines I was able to find (with silicon assistance): 17.Bxh7+ Kf8 [17...Kxh7 18.Nxg5+ Kg8 19.Qh4 Ne7 20.Qh7+ Kf8 21.Qh8+ Ng8 22.Nh7+ Ke7 23.Qxg8 Bb5 24.Nd5+ Kd7 (24...exd5 25.e6 Rd7 26.exf7+ Be2 27.f8Q+ Kd8 28.Qxe8#) 25.Nxc7 ] 18.fxg5 dxe5 19.Nh4 Nd6 20.Ng6+, etc. (as actually played).