Why stockfish cant find the solution?

here is the puzzle white to play and win:

lichess.org/7fzCRPFe64aE

solution:
lichess.org/study/FIJicgr8
lichess.org/video/ncdsvKP0pPA?tags=tactics

Why a.i. cant find the right solution? Why the move Bc2+ change so much the evaluation to -6.8 -> +4.3?

You made a mistake setting up the position. The knight should be on a6, not a7.

hahaha mistake is now fixed

Wow that's an amazing puzzle.

Still the wrong position

I think there should be a black bishop

now its really fixed!

You can read about this puzzle here:
gameknot.com/room-fm.pl?home=2&th=1090

"This is a chess problem steeped in legend.

Imagine if you will, the year is 1987; the location, Brussels. Some of the strongest players in the world, including recently crowned world champion Kasparov, are participating in one of the biggest tournaments of the year. Even the world champ must work very hard just to keep the pace with Karpov and Ljubojevic, both on top of their games. Also playing are famous grandmasters such as Jan Timman, Viktor Korchnoi, Bent Larsen, Nigel Short, and the living legend Mikhail Tal.

Darting in and out of the press room are all manner of GMs. Suddenly, in pops Jim Plaskett, a young and intense British grandmaster. In between telling jokes and talking politics, he sets up a most intriguing position. "White to play and win" he pronounces, then departs, leaving the position as a parting gift. Or perhaps one should say a parting shot!

For several hours, chess fans and GMs alike try their hand at unraveling Plaskett's mysterious puzzle. While some of them apparently make progress in the analysis, no one can completely work out the complexities of this deceptively deep position. Even the computers choke on it.

The tournament was a tie between Ljubojevic and Kasparov (both with 8.5/11), but Misha proved his genius in a way that the other GMs could not. After struggling with the puzzle for a few minutes, Tal left the playing hall to take a walk. An hour later, after some fresh air, the Wizard from Riga popped back in and immediately played the correct solution on the board, a variation so subtle and beautiful that it stunned all who witnessed it.

So what is the true origin of this puzzle? The best we can determine, it was invented by the Dutch composer Gijs van Breukelen sometime around 1970. Instead of publishing it, he merely showed it to some friends. Being so incredible, it found its way through the grapevine into grandmaster circles. By word of mouth it found its way to Spassky, and later to Plaskett. In 1997, Breukelen finally decided to publish it, in the Dutch chess magazine Schakend Nederland.

Can you solve the puzzle that stumped a room full of grandmasters? Take your time. "

This is about a famous endgame study, involving GM Mikhail Tal solving it where other GMs (Karpov, Kasparov included) failed to do so.

www.youtube.com/watch?v=BUk_zWrbcsM&feature=youtu.be

To directly answer the question, engines have a hard time with this position precisely because of some of the features that make them so strong.

Even for engines, searching the tree of all move sequences exhaustively is just infeasible; search depths would be very low and play would be very weak.

Modern engines are as strong as they are in large part due to clever tricks that programmers have come up with that let the engines search more deeply in promising lines while searching much less deeply in lines that are probably irrelevant.

When a modern engine tells you that it's reached "depth" 30, for example, that actually just means it's completed 30 iterations of its search.

The mainline will have been searched more deeply than 30 plies (a move by white is 1 ply, black's response is a 2nd ply, and so on), while a lot of lines will have been searched to much lower depths.

One example of these tricks is to heavily reduce lines where one side is down a lot of material and playing quiet moves. In the vast majority of positions this makes a great deal of sense. If you're down a queen, quiet maneuvering is a dead-end; you'll look to see if you have winning checks, captures, mate threats, etc., but in 99.9999% (a made-up number that's probably not too far off) of positions if you're down a queen and making quiet moves, you're just lost and it would be a waste of time to keep calculating those lines.

With this puzzle, we run into a position from that 0.001%. White is down substantial material to start, and the solution involves giving away an additional piece, throwing away our only apparent trump (the passed pawn), and after all that making a series of non-capture, non-checking moves (Ke6, Ke5, Be4, and the resulting bishop dance).

From the start position, that line will just get reduced very heavily, so Stockfish doesn't see deeply enough into the line to realize the mate threats are there.

Once you force some of the heavily reduced moves on the board, SF can't reduce the lines as probably irrelevant (since they're actually on the board!), and since it's searching them to full depth now, quickly sees the mate.

It just illustrates a truth about chess. It's a very, very complicated game, and the edge cases can get very strange very quickly.

Approaches to the game that help you play the vast majority of positions you see in tournament play can fail to solve these weird positions, and approaches that work well for these weird positions don't work that well in the more "normal" positions (as an example, there are versions of engines that are designed to be better at solving these weird sorts of study positions, and they are invariably substantially weaker in game play, even though they solve these positions much more quickly).

What a game, huh? :)