So, I see 33...Qe4+, which forces the white K to a square (f2, g1 or h2) where we can deliver check with our knight -- which is good because we badly want to move our knight (if for no other reason) to vacate e5, clearing the path for both our rook and our bishop.

So, we just have to check out those 3 options for white:

If 34.Kf2, then 34...Nd3+ 35.Kg1 Qe3+ 35.Kh1 (if anything else, 35...Qxg3+ ~#) Nf2+ 36.Rxf2 Qe1+

If 34.Kg1, then 34...Nf3+ 35.Rxf3 (if king move, then 35...Nh4++ & 36...Qg2#) Qe2+ and sooner or later, we bring the rook in at e2 or e1 with check, or play Qxg3 with check, winning.

If 35.Kh2, then it's pretty much the same as 34.Kg1.

Materially, white is a little better since they have the bishop pair. White threatens Qh6 and mate on g7 or h8. But it looks like black can win a piece with Qxd3. Let's see if it works:

32...Qxd3 33.Qh6 This seems to be the only forcing move to consider for white, since 33.Qxd3 Nxd3 just concedes a piece.

Now 33...Qxg3+ 34.Kxg3 Ng4+ 35.(K moves) Nxh6 wins a piece and a pawn. And if after 33...Qxg3+ 34.Kh1, black has at least 34...Qxh3+ 35.Kg1 Qh2+ 36.Kxh2 Ng4+ and Nxh6--not that this is necessarily best. OTB, this line is already winning so analyzing further would be a waste of precious clock time. When move 33... arrives, I would consider Re2+ to see if it's better than the Qxg3+ line.

I didn't like Qe4+ because as far as I could see black wouldn't be able to immediately checkmate white whereas Qxg3+ is the safe solution. But I dismissed Qe4+ too early I see.

32.Bf6 is attacking; obviously white is thinking 33.Qh6 and 34.Qh8++. Black must respond quickly and accurately as the threat is real. I dont think attacking here is "getting cute", but more of a miscalculation on white's part.

If I didnt know this was a puzzle I would probably make the incorrect move 32...Kh7 and miss the counterattack. White is hanging a piece, afterall, and is threatening mate! Why would white hang a piece if he didnt see something more important...like mate? Psychologically, it would be hard to sit there facing Qh6 and start calculating lines of counterattack. To Black's credit, he did.

What sweet words you wrote:

"33 ... Qe4 + is way better."

You are right about the gravity of the threats. (There was also a pin on black's knight).

Speaking for myself, I saw 32...Qxd3 right away, and if 33 Qh6, then 33...Qe4+ looked good.

I did not see every move of the mating sequence, but with white's queen unable to help after 33 Qh6, I hope I would have played 33...Qe4+ instinctively.

A) 33.Qxd3 Nxd3 and Black is a knight ahead.

B) 33.Qh6 Qxg3+

B.1) 34.Kxg3 Ng4+

B.1.a) 35.K moves (or 35.Rf4) Nxh6 with N+P ahead.

B.1.b) 35.Be5 Bxe5+ 36.Qf4 (otherwise 36... Nxh6) Bxf4+, again N+P up.

B.1.c) 35.Qf4 Nxf6 36.Qxd6 (otherwise 36... Bxf4) Ne4+ and Nxd6, again N+P up.

B.2) 34.Kh1 Qxh3+ 35.Kg1 Qg3+ 36.Kh1 Qh2+ 37.Kxh2 Ng4+ and Nxh6, now N+2P ahead.

Interesting. In a lot of these chessgames puzzles the side to move usually has some shocker move that causes the other side to react and to try and mount the best defence possible (sometimes there are surprisingly good defences missed by the defending side and by the chessgames puzzle pickers).

This case is a little different. Black can pick up a piece with 32 ...Qxd3. But this allows 33. Qh6 and white threatens mate next move. Seeing this I have to admit that I looked at a lot of alternative 32 moves for black: ...Nf3 ...Ng4 ...Nxc4 ...Nxd3 ...Bf8 ...Kh7

I wasn't really very happy with any of those alternatives (some of course for fairly obvious reasons). Actually 32 ...Kh7 is not a horrible move (I think white's best reply is 33. Re1) but clearly I missed the whole point of the puzzle.

I then gave up and peeked at the actual move 32 as played in the game. After seeing the intial move was indeed 32 ...Qxd3 I then thought about how the rest of the game must have unfolded - but not with much effort I must admit.

My mistake was obviously to have given up on 32 ...Qxd3 too quickly. For some reason I didn't bother to try and see if black had enough forcing moves to win this one. Still I am not sure that I could have calculated this one out accurately, even if I was convinced from the outset that ...Qxd3 was the correct move. I am pretty sure I would have played 32 ...Kh7 in a real game situation.

Before playing 32...Qxd3 and seeing the strong threat 33.Qh6 I would "test" the line 33...Qxg3+ first and seeing that it is winning I would go ahead and play 32...Qxd3 and hit the clock. After white actually plays 33.Qh6 and if I have plenty of time on the clock then I may spend time looking at moves like Qe4+.

I spoke with NM Dan Heisman on this and the first thing he analyzed was the Qxg3+ line. He said IM Donald Byrne showed him a position years ago and Byrne asked him what he was thinking about because a certain line won a piece. Dan said he was trying to see if there was a forced mate, and Byrne said "Don't! With the extra piece and as good of a player as you are, any grandmaster would resign against you."

Dan said that after a game Kramnik was asked why he didn't play another line for a shorter win and he replied he saw a winning line and there was no need to look at it further.

Dan also said if he had 7 minutes left on the clock he would go ahead and play Qxg3+ and win the endgame easily in those 7 minutes. But if he had 2 hours left on the clock, he would spend time looking at Qe4+ to see if it's better.