Probably just because the evaluation by the engine drops a little and, according to evaluation you had some better move. Even if your move is still winning.
Why is this a blunder?
This reminded me of how i finished my last online game (960) when i managed to be a Rook up:
Probably a program would find blunders/mistakes in those moves, too, because i gave back some material to my opponent, and then refused to take a pawn, but in reality they got me from a winning middlegame to a winning endgame, in other words they secured victory for me, which is hardly insignificant.
I think this is another reason why we shouldn't depend on computers to analyze our games for us. They may be able to look further ahead than even most gms, but other times they miss obvious positional play.
If you're referring to the chess.com analysis everything that isn't perfectly played is a blunder, error or mistake - according to the chess.com analysis. It's also very NON-helpful in pointing out that you "erred" when you were already 99 pawns down because with "best play" you would only have been 77 pawns down, or vice versa - ie you were 88 pawns up and now after you "erred" you're only 66 pawns up.
OK, so I'm exaggerating...but not by much
PS: "It" also will state on occasion a perfectly sound opening variation is "inferior" or a "mistake" so don't take it's comments on opening theory too seriously either
Rxf3 is far better than complicated the game with B vs Rook endgame. By capturing the Bishop, Black ensure he will end up 2 pawn up which +2 already and potential queen +9. The other player should realize the pawn endgame is much simpler and way to go. The engine correctly analyse it as the best continuation and even me would go for it. That why endgame is very crucial for any chess player. Chessbase endgame product from GM Karsten Muller and GM Smirnov's endgame course is recommended. Chess book , Silman Complete Endgame Course would be my choice
download Free Video how to prevent blunder by GM Smirnov : http://gmigorsmirnov.blogspot.com/2012/11/get-free-videos.html
Just fyi-- computers can't do endgames.
With endgame tablebases they do suprisingly well
This endgame is from a game I played quite recently. It was interesting so I submitted it for analysis. The analysis came back saying that ...Rxf3?? was a blunder but I cannot see how, even looking at the alternate lines the analysis provides.
The way I see it, assuming the king captures the rook on f3, kg5 wins white's last pawn, leaving it a clearly won 2pawn vs king endgame. In addition, I played this endgame against a chess engine I have and as soon as I played Rxf3, the computer resigned as white. In my opinion, Rxf3 guarantees black to queen a pawn. Maybe its not the fastest to win but it makes winning easy, its hard to screw up a 2 pawn vs king endgame. In the actual game the other player then went for my f-pawn and got it but I easily queened my h-pawn and won. So why is the analysis engine not picking up on this?