Sometime last month IM and computer scientist (and regular reader of this blog) Ken Regan found a little time for tournament chess, and while the result wasn't all he might have hoped he was kind enough to send one of his games, from which I'll present the following excerpt.
IM Kenneth Regan - NM Scott Riester, position after 20...Qxh2+ 21.Kf1
"Now the pin-cutting 21...Bc5! was missed by both players. [DM: That very nice move is best and conclusive, but Black has other moves that also win with a huge margin.] The computer sees instantly that it's "splat!", so my statistical model rates it a near-certainty to be found, but this may qualify as a kind of "Invisible Move" that is hard to suspect let alone see. My opponent's actual 21...Qh1+ 22.Ke2 Qxg2 seemed natural, and now I hallucinated that after 23.Nxa7+ Kd7 24.Rg1 Qe4 I could play 25.Rgd1+ Ke7 26.Bg5 "mate". Seeing nothing else I played 25.Qb5+ and resigned next move. But I could have really mixed things up by playing 23.Bxf5!
"According to my Houdini run to high depth, there is only one way for Black to keep the advantage, let alone survive. Can you find the move, unaided?"
I (DM) will give the answer later, in the unlikely event that someone doesn't find it first and include it in the comments. But please, readers, only give what you find with your own thoughts; don't include or hint the solution if you got it from an engine - even if you tried to find it on your own first!