A tactical puzzle that Stockfish missed.

8/1p1pNpbk/1q1P4/pP2p2K/P3N3/4P1P1/3P4/8 w - -

Solution. 1. Nc8 Qe8 2. Ng5+ Kh8 3. Nf7+

It didn’t fail because Nc8 Qe8 Ng5+ Kg8 and now there’s no fork 

If you use stockfish, SF could not find the solution ( the winning move) for a long time.

Instead he will stack with 1. Ng5+ , 0.00 (draw).

Nc8 Qd8 Ng5+ Kg8 Ne7+ Kf8 Nh7+ Ke8 Nd5 Qb8 Nc7+ Kd8 b6 Kc8 g4 e4 Ng5 Bf8 Nxf7 Bxd6 Nb5 Bc5 Na7+ Qxa7

Yeah, a "simple" 12 move queen trap involving domination and zugzwang, so "easy."

But it is surprising an engine fails a puzzle that involves many forcing moves.

Obviously it prunes away that tree before going deep enough.

Weird

That solution was done by Sting engine version 11.2.

I agreed, it was not easy. hehe.

I didnt check all those engine lines before I post. I thought the answer would be 3-5 moves.

This could easily be a 3500+ type of puzzle.

(Some) humans would be able to solve it though.

No, @2012vachamp who uses an engine on TT wouldn't be able to solve it, because like we said, engines can't solve it lol

Doesn't matter what black plays on move 8 because IIRC that endgame is winning after the black king and queen are trapped.

Just treat it like a N vs B endgame where you're basically up a king.

Just go download stockfish and run it for yourself.

Not chess.com's java junk stockfish, get actual stockfish.

Anyway:

Yeah, at least at low depth, my SF isn't seeing it until move 6 or 7.

Yes engines are good at calculation, but if calculation was all they did, they'd still be weaker than GMs. Chess is far too complex to solve (or be any good at) with brute force.

So a big part of engine strength is their search functions that essentially ignore lines that don't seem promising. So in this case engines decide the winning line is bad and never look at it deeply enough to realize it's winning.

You can probably get around it by messing with settings on how it prunes trees. I remember Pfren saying his engine would solve a puzzle like this very quickly after he adjusted certain settings.

i think stockfish makes its moves based on win percentage, and in the initial moves, white has a low win percentage

I guess you mean Monte Carlo tree search? It doesn't use that.

ok didn't know, thanks for telling me though!