Stockfish Blunders.

White wouldn't even get the chance to checkmate because Rxb2 would be check and after the king's only legal move (a1), you'd do Rxa2+, followed by Rxb2#. Wouldn't black win?

hi

Ok, I assume this is the original position and it's black's turn to move:

I ran Stockfish 13 on 12 cores in one of my pcs. These are the 3 best lines:

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "6k1/5R1p/2p3p1/p1qp4/3Q4/2P5/PP2r1PP/1r3RK1 b - - 0 1"]
[PlyCount "60"]

{Stockfish 13: 1)} 1... Qxd4+ 2. cxd4 Rbxb2 3. R1f2 Rxf2 4. Rxf2 Rb1+ 5. Rf1 Rxf1+ 6. Kxf1 Kf7 7. Ke2 Ke7 8. h4 Ke6 9. Kf2 Kd6 10. Kf3 Kc7 11. g4 c5 12. dxc5 Kc6 13. Ke3 Kxc5 14. Kd3 a4 15. Kc3 d4+ 16. Kd2 Kd5 17. Ke2 Ke4 18. a3 Kd5 19. Kd1 Kc4 20. Kc2 h6 21. g5 h5 22. Kd2 Kd5 23. Kd1 Ke5 24. Ke1 Kf5 25.
Kd2 Kf4 26. Kd3 Kg4 27. Kxd4 Kxh4 28. Kc3 Kg4 29. Kd3 Kg3 30. Ke2 Kg2 31. Ke1 {-7.11/39} *

({Stockfish 13: 2)} 1... Rxf1+ 2. Rxf1 Qxd4+ 3.
cxd4 Rxb2 4. Rf6 Rxa2 5. Rxc6 a4 6. h3 a3 7. Kh2 Rd2 8. Ra6 a2 9. Kg3 Kf7 10.
h4 Ke7 11. Kf3 Kd7 12. g3 Kc7 13. Ke3 Kb7 14. Ra3 Rh2 15. Kf4 Kc6 16. Ke5 Re2+
17. Kf6 Kb5 18. Kg7 h5 19. Ra8 Rc2 20. Kh6 Kc4 21. Kxg6 Kxd4 22. Kxh5 Ke3 23.
Ra3+ Ke4 24. Kg6 d4 25. h5 d3 {-4.34/39})

({Stockfish 13: 3)} 1... Rxg2+ 2.Kxg2 Qxd4 3. cxd4 Rxb2+ 4. Kg3 Rc2 5. Rc7 Rb2 6. Rff7 h6 7. Rfe7 Rb8 8. Rg7+ Kh8 9. Rh7+ Kg8 10. Rcg7+ Kf8 11. Ra7 Kg8 12. Rxh6 c5 13. Rxg6+ Kf8 14. Rh6 Kg8
15. dxc5 Rd8 16. Kg4 Rc8 17. c6 a4 18. Rg6+ Kf8 19. Rd6 Rb8 20. Kg5 d4 21. Rxd4 Rc8 22. Rd3 a3 23. Rd4 Kg8 {50.06/39})

Rbxb2 happens AFTER the Queen trade. And yes, Rbxb2 in the position from my post above is mate in 5. 

1... Rbxb2 {Stockfish 13:} 2. Rg7+ Kh8 3. Rf8+ Qxf8 4. Rxg6+ Re5 5. Qxe5+ Qg7
6. Qxg7# {#5/5} *

Sorry position in the image is wrong, rook was on b1. Anyway, the main line still is 

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "New game"]
[Black "?"]
[Result "*"]
[SetUp "1"]
[FEN "1r4k1/5R1p/2p3p1/p1qp4/3Q4/2P5/PP2r1PP/5RK1 b - - 0 1"]
[PlyCount "52"]

{Stockfish 13:} 1... Qxd4+ 2. cxd4 Rbxb2 3. R7f2 Rxf2 4. Rxf2 Rb1+ 5. Rf1 Rxf1+
6. Kxf1 Kf7 7. Ke2 Ke7 8. h4 Kd6 9. g4 c5 10. dxc5+ Kxc5 11. Kd3 d4 12. a4 h6
13. g5 h5 14. Kd2 Kb4 15. Kd3 Kxa4 16. Kxd4 Kb3 17. Ke5 a4 18. Kf6 a3 19. Kxg6
a2 20. Kh7 a1=Q 21. g6 Qb1 22. Kh6 Qc1+ 23. Kh7 Qc2 24. Kh6 Qd2+ 25. Kh7 Qd3
26. Kh6 Qd8 27. g7 {-5.82/30} *

Sorry, The R is on b8.  

The black Rook was on b8, not b1.  I guess that would change things.  I commonly forget the numbers run backwards for the black pieces.  To me, my back rank is the first rank whether I'm black or white.

As I mentioned above, I tried both positions - rook on b1 and b8, and the Rbxb2 is not the first engine movie in the position. The engine exchanges queens first and then goes Rbxb2, because otherwise it's mate in 5. So I dont understand why it'd say Rbxb2 is a good move. A bug maybe? No engine would call it even at a very low depth.

idk why Stockfish blundered. I guess we can learn that: even computers make mistakes.

I run my Stockfish 13 NNUE in my android ( droidfish) and Stockfish see Qxd4 instantly. ( within 1 second)

If people dont know how to get Stockfish 13 in android, here is my post. 

https://www.chess.com/forum/view/general/how-to-install-stockfosh-13-in-android-phone

Yes computers can blunder if you throttle to 1/1000 of cpu resource or run equavilent of 0.001 cpu time in a complicated position.  

> The computer analysis says Rbxb2 is a good move.  In a previous analysis it said it was the best move.

Here's the FEN for anyone interested:

1r4k1/5R1p/2p3p1/p1qp4/3Q4/2P5/PP2r1PP/5RK1 b - - 0 1

I tested this on:

1. Stockfish 15
2. Stockfish 14.1
3. Stockfish 12
4. Komodo 12.1.1
5. Lc0 v0.27.0
6. Igel 3.0.10
7. Maia 1100 (just for fun)

All of them report Qxd4+ (c5d4) at depth 1 (and beyond, so there's no rogue Rbxb2 sneaking in on depth 12, for example).

Side note: I used Game Review on a Wesley So game featured on one of the Opening pages. It reported one of his moves as a blunder. After switching to the Analysis tab (Stockfish in your browser) the eval changed from Blunder to Best Move, and what they had as Best Move changed to a horrific Blunder. I believe they update the saved analysis on the backend so it probably wouldn't be repeatable, but I have a screenshot of the before-and-after.

I've checked my own games against local Stockfish, to say the least I can't figure out which version of Stockfish they're using or at what depth. I believe resource constraints have them getting creative and the seems are starting to show. The Analysis tab is legit, but Game Review gets a ?! from me for the time being. Unfortunate because I really like the concept and the UI they've created.

OTOH if you really wanted to analyze every move in a game to depth 30 (fixed depth isn't even a good technique, but it's easy to understand), especially in your browser, I think you'd be just as unhappy but for different reasons. Mainly that you'd wait an hour or more. "Other sites" have problems with server-side analysis too, but when you call something "Best Move" I think there should be reasonable confidence that assessment is close to accurate.

Wow thanks for the info.  I am in love with the interface.  I wish I could use an engine on my computer with it.  I have one, but it makes me cross-eyed trying to follow the lines.

Not sure anyone tried this, but you can change the depth for the game review. It’ll take more time, but it’ll also make more sense. I guess it may depend on what membership you have, how much cpu time they grant. 

I could test, but there’s no full game dump here, just the FEN, which shows the correct moves right away.

I've used the deep search button on occasion.  It hasn't helped.  

I'd like to use an engine on my computer with the interface.  The interface makes the rehash bearable, but Stockfish can be sketchy.