This Position BREAKS Stockfish

I found a position that breaks stockfish:
lichess.org/analysis/1k4rr/2b1p3/3pPp2/2pP1Pp1/1pP3Pp/pP5P/P7/4K3_w_-_-_0_1

Here the engine estimates a -15.8 in black's favor, but the game is a forced draw because there is no possible way for black to force a pawn breakthrough. White can keep moving his/her king for the rest of the game and can force a draw by the 75 move-rule (however, stockfish can't see this...). You can give black as many rooks and dark-squared bishops, but the game will still be a draw. White does not have to, and should not accept any of blacks' sacrifices, as that will lead to a possible win for black.

That's one of the biggest flaws of chess engines imo. They think materialistic not positional. I hope AI's will make a better job at these positions.

I played over the position practically with stockfish 10 at depth 18, and only after white declines the sacrificed rook at move 46 does the engine realise the error of its ways, crashing the evaluation back to 0.0.

It's good to look at these positions and see how the traditional chess engine fails to properly evaluate the position. Let me also turn your attention to this classic gem of a starting position along the same idea: lichess.org/KYaZzbSH

Stockfish gives a huge material advantage to Black even though it is not nearly that simple; Black's pieces are completely trapped on the first and second ranks.

Its kinda funny tho..
Makes u feel smart compared to a computer ;)

Why not report this to the stockfish devel team ? This may help improve the engine ...

Way to go. It breaks leela also. leelanalysis.com/

This is a -99.0 but a draw.

it is impossible to checkmate white
same could happen with the white bishop army
very funny tho :)

Stockfish is evaling the position is better for Black, as that’s what it seems. But because the eval is so good Stockfish feels confident there is a win even though it hasn’t been concisely calculated out. (My guess anyways)