Stockfish 11+ not seeing mate in 6?

I imported the following OTB game I played with a friend.
The evaluation of 13. Qb2 should be #-6. There is nothing stopping the queen marching Qg5 Qg4 Qh3 Qg2#.
If I manually move Qg4 out of the main line and against the recommended best move, only then the analysis yields #-4.

I tried to give the engine some time and let it "go deeper", but the best it can find is some -35 evaluation.

Not sure if it is a bug or I made some miscalculation. Happy to be proven wrong.

The problem here is that you didn't let it run to a high enough depth. It finds it at depth 19 for me.

@IndianDefense said in #2:
> The problem here is that you didn't let it run to a high enough depth. It finds it at depth 19 for me.
I suspect you were using SF 14 in the lichess analysis board

SF 11 will find this mate, although it does have to go to a higher depth.
imgur.com/a/XNQHuMg

With SF11, I did have to run to depth 28 to announce mate in 6 (about 500M nodes).

@IndianDefense I swear I let it run some good minutes and tried reloading the site a couple of times. But good to see it works fine for someone else. I was using an old PC with poor performance, might be that. Thanks for trying that out.

No, I turned off NNUE and ran Stockfish 11+ HCE. It found the mate at depth 19. Memory was set to 512 MB if that matters.

@corvusmellori said in #3:
> I suspect you were using SF 14 in the lichess analysis board
>
> SF 11 will find this mate, although it does have to go to a higher depth.
> imgur.com/a/XNQHuMg
>
> With SF11, I did have to run to depth 28 to announce mate in 6 (about 500M nodes).

I believe for me I went down to depth 32. Yet as I wrote above it might be that the PC I was using had quite poor performance. Thanks for double checking.

@Giuiba said in #6:
> I believe for me I went down to depth 32. Yet as I wrote above it might be that the PC I was using had quite poor performance. Thanks for double checking.

It will also depend on your memory hash, threads you're using, etc. For example, I used one thread (instead of 28 threads) and SF11 announced mate in 6 at depth 33 (after about 300M nodes).

Stockfish's main purpose isn't to evaluate positions, and it's not even to find the best move; it's to win games. Even at depth *two*, it suggests Bxc3, which while not doing a lot, is still definitely a winning position for Black. At depth 20, it's suggesting Qg5 as the best move.

There are some positions even the most recent Stockfish doesn't evaluate correctly, but it's mostly the kinds of positions that come up rarely and don't impede its ability to win chess games.