And if the rook wasn't there, Nxc6 forks the black queen and rook.

So let's doodle with some "mad" moves. 22. Qxb7 Bxb7 doesn't work. 22. Qxc6 and 22. Nxc6 are also non-starters. So let's try 22. Qxa6. This deflects the b7 pawn away from the defence of Nc6. And once we have exchanged knight for d8 rook, we can snatch a pawn with Bxa5.

I didn't spot 24. Nxc3 but it does not worry me in the slightest. It opens the diagonal of the g2 bishop and accelerates the doubling of the rooks on the c file. White is up by two rooks for a queen in a position with open files - and that is good enough for me.

22. Qxa6 bxa6 23. Nxc6 Qd7 24. Nxd8 Qxd8 25. Bxa5

Material: B for N. White has an outpost Ne5 supported by Pd4 exerting pressure on the Q-side light squares with Bg2 and Qb5. The White Rc1 and Rd1 are connected and actively placed controlling the center. Overall, the Black position (particularly Bc8 and the back rank) is cramped but relatively solid. The Black Kg8 is lightly defended. There are no obvious candidates, so on to the usual suspects: no checks, but several captures are possible. Only Bc3 requires activation.

Candidates (22.): Bxd5, Nxc6, b4, Nc4, Be1, Bxa5

Instead, after 22. Qxa6! Nxc3,

click for larger view

White should play 23. Rxc3! (not 23. Qd3? Nxd1=), when Black has no good alternative to 23...bxa6 (23...Nxe5 24. Qxa5 ) 24. Nxc6 Qd7 25. Nxc8 Qxd8 26. Rdc1! (transposing back to the winning game continuation).

So instead of one really good idea, I had for two rather tame ideas:

Tame idea #1: 22.Nc4, putting extra heat on Pa5. How can black defend? I suspect he would try 22...Qc7, and then comes 23.Be1, which begins to open the c-file with pinning threats on the black knight. I threaten to open it further with 24.Ne3 with lots of heat on d5 and c6, or if 23...Na7, then 24.Qc5. I think white might just be better!

Tame idea #2: 22.Be4, and my bishop is nicely developed, containing traces of a distant kingside attack!

I have to admit that delaying Nxc3 never occured to me (it IS very weak so that might explain why), neither did the final blow Bb7!, although I do not consider that as vital for the combination as the first three moves: as 2R vs Q is usually quite an advantage anyway (I have to admit that in Blitz I prefer the Queen, but I won a two Rooks vs Queen endgame yesterday nonetheless).

-ML-
<22.Qxa6!> not too hard seeing that the White Queen is quite locked up into the Black position, and this is at this point the only way to really get anywhere for White (retreating with Qb3 being drawish). I not gonna to vary too much today: if Black does not capture the Queen, White will simply win AT LEAST the exchange.

<22....bxa6 23.Nxc6> as Pb7 was its only protector, and now White continues with the Knight fork winning an exchange and thus ♗+♖+♖ vs ♗+♕ with a better position. However, Black should still watch out how to react.

<23....Qe8> this doesn't really matter, Black should just defend d8.

<24.Nxd8> now Black can vary a bit, which I did not consider tbh.

/A\
<24....Nxc3> with some analysis the wrong way to go, but it seems clear that White already gets his advantage of two pieces versus one one move later:

<25.Rxc3 Qxd8 26.Rdc1 > okay, in the puzzle I missed the forced continuation 26....Bd7 27.Bb7! with 28.Rc8!. Still, even without this tactic, White is winning after move 26, and I would've continued like this OTB: once the position is there after move 25, it's not hard to spot.

/B\
<24....Qxd8> the better alternative seeing the winning line after Nxc3, but now White can still use his ds Bishop, and do this exactly one move later.

What about 22. Qxa6, Nxc3 23. Nxc6,bxc6 24. Qxa5,Nxd1

The Master Chess (32 bit 20 MHz) evaluates this as +0.7 so white has an edge, but not a decisive advantage!

Could somebody check that line for a mistake / improvement?

[1] Agreed.