Stockfish wrongly evaluates this position

It is a draw, but it says it is winning for black

Black can sack a rook and break through white's pawn formation. After all, he is up in material

Nvm You don't have to take

Stockfish cant calculate fortresses. Neither can many other computers

@TCF_Namelecc Yeah it is interesting i tought the new version recognizes this

What is sad is that if you played Rb5, it recommends that white take the rook xD

@TCF_Namelecc Hahahha That's ridiculous! That's why never 100% trust an engine

Some studies which have this as topic:

lichess.org/analysis/rkb5/1p6/pBp5/PPPp4/1K1Pp1P1/4Pp1p/5P1P/8_w_-_-_0_1 White draws
lichess.org/analysis/8/kp2q3/1Np1b3/pPPp4/P2Pp1p1/4Pp1p/5PPB/1R4K1_w_-_-_0_1 White draws
lichess.org/analysis/2qN4/k2p4/2pP4/1pP3B1/pP6/P3K3/8/8_w_-_-_0_1 White wins

In all these studies Computers give wrong evals and can not find the solution. At least for a long while (didnt check).

lichess.org/analysis/8/7n/8/4k1p1/5pP1/4pPRB/2BpPRQB/3KRQQQ_b_-_-_0_1

This one is interesting, Stockfish 9 doesnt understand this and gives around +60, but it is a draw. Meanwhile Stockfish correctly evaluates this position with 0.0.

lichess.org/analysis/k1bB1B1B/2b1B1B1/8/4B3/8/1n1p2P1/4N1PB/3KRBRQ_b_-_-_0_1

This one is also interesting. If you remove the white bishops on the eight and seventh rank stockfish gives 0.0 (the correct eval). But with the bishops Stockfish is not able to see that it is draw.