strange evaluation line by stockfish

in 23 move, stockfish sac his queen for the attack, I am not sure if it is a blunder

It seems counter intuitive at first glance but everything that happened seemed relatively forcing from an engines point of view I guess. Hard to call it a blunder when they won handily against another engine

An engine does not blunder.
It is remarkable.
Somewhere in the calculations there is a justification, but is is hard for humans to see.

Note that it's a queen for R+N+p, not exactly a large sacrifice. And right after that it seemingly sacrifies a minor by not retaking at c5, and right after that white has to sac an additional minor itself against b3, to stop the Na3 attack. Then, SF has the option of winning back the queen for the R+N (with 33... Rb1+), but doesn't take it.

Very much not a blunder, but rather... a lot of foresight.

stockfish evaluate:
position fen rqb2rk1/2p1ppb1/2Np2p1/2nP2Np/ppP1PPPP/3nB3/PP4Q1/1K1B3R b - - 3 23
go nodes 3000000000

info depth 47
seldepth 35
multipv 1
score cp 0
nodes 3000000916
nps 538715
hashfull 1000
tbhits 0
time 5568804
pv b8b6 g4h5 d3b2 d1e2 c8a6 h1c1 a4a3 e3c5 b6c5 c6e7 g8h8 h5g6 b2c4 e2c4 a6c4 c1c4 c5c4 g2f3 g7h6 f3h5 c4d3 b1c1 d3e3 c1c2 e3f2 c2d3 f2f1 d3c2
bestmove b8b6
ponder g4h5

Overall: it is weird. Stockfish eval is 0.00

It's insane how complex and unnatural the positions these computers get into. Can you imagine a human player getting to this position?

I find out stockfish make another evaluation mistake in move 25, stockfish think Be2 is the best move, but, if white play Be2 in move 25, this can happen. it is a crazy one: lichess.org/lPb3CeVF/black

I like how that queen casually comes back from the dead. Don't try this at home folks...

@onlyplaytired Actually the game bears a strong resemblance to a certain line in the Sämisch King's Indian: lichess.org/JGQlvb63#21
8...Na5 is rare and 9. Nf4 is the rarest (but sharpest!) response to it.