A curious mismatch between “human eval” and Stockfish

Human understanding of chess situations is fallible, yes, but this example from an over-the-board game I played yesterday (with a player rated 500 more than me) takes the cake.

I lost. After the game, it was absolutely clear to both of us where I had blundered and how his move had sealed it. But when I analyzed using Stockfish, his greatest move, so-obviously-the-best to human intuition, was actually a blunder!

I am playing White here, and here is a situation where I am winning, with an eval of a humungous 472 in my favor — — that is like being up up by 5 pawns.

My greatest weakness is when I think I am ahead, I slack off and lose. Here is the situation after a few more moves, just before his killer move. I have just moved the queen. Black to play. Can you spot Black’s killer move?

Yup, Bc8 skewers my Queen and Rook. My Rook is as good as dead. What an obviously brilliant move!

But stockfish disagrees. It says: before Black’s move, Black was winning (eval 291 in favor of Black). But no longer! The eval after his move is an exact 0, it is a draw. I wish I had figured it out then, in the UCSF rated tournament, I certainly wouldn’t mind a draw against a player rater 500 more.

What I should have done here: 24. Rxh7+ Qxh7 (forced) 25. Qxf6+

Now Black cannot block check with Rook, since 25…Rg7 looses his other rook to 26. Qxd8+. If he blocks check with Qg7, I play 26. Qh4+, a perpetual check!

Moral, as always: “Calculate!”