I had this position arise in a game:
[FEN "4qrk1/3rppbp/1p4p1/p2P4/1pBP1B2/1P2PN2/5PPP/R2Q1RK1 w - - 2 21"]
[White "Kevin"]
[Black "lichess AI level 3"]
1. Ne5 {I advance a knight and, incidentally, attack the rook. In retrospect, this move looks just plain wrong to me, because Bb5 is better, but Stockfish says it's OK.} (1. Bb5 {White wins a rook for a bishop immediately and Black has no obvious counterplay aside from trying to win the pawn on d5. This is the best move according to Stockfish.}) Rc7 {Stockfish says this move is fine but I'm not sure why. The rook has at least two other escape squares, and this allows a discovered attack.} (1... Ra7 2. Nc6 Rd7 3. Ne5 Ra7 {White doesn't want to repeat moves because they are winning, so now White's attack is temporarily stymied. Stockfish suggests Qf2 and further developing moves.}) (1... Rb7 2. Qf3 Rc7 {This is the Stockfish line following Rb7. It seems that Stockfish doesn't like the rook on b7, probably because d6 is strong if the rook stays put.}) 2. Nxg6 {Fork number 1.} e5 {Refuted!} 3. Nxe5 (3. dxe6 {The wildest line I saw in my entire analysis. I failed to consider this at the time because I forgot that the rook on c7 is still attacked by the bishop, but frankly I find it hard to believe that I could have found all of these moves.} Rxc4 {The rook is hanging, so it had to move somewhere, but I think we sacrifice it to prevent lines involving exf7+.} 4.e7 {And then White just goes after the other rook instead.} Rc6 5.exf8=Q+ Bxf8 6.Ne5 {And the knight escapes!}) Qe7 {Again, Stockfish thinks that Qe7 is an OK-but-not-best move, but I don't see why the queen even should move at all, let alone to e7.} (3... Rc8 4. Qh5 {Stockfish line. From here, Stockfish makes a lot of mating threats and generally plays like it's an endgame, despite the fact that both queens and all four rooks are still on the board.}) 4. Ng6 {Fork number 2.} hxg6 {Not a direct refutation, but we don't want to sacrifice two minor pieces for one rook.} 5. d6 {Fork number 3, and this one finally sticks.}
I missed Bb5, and in retrospect it is clearly the best move in this position, both by my own analysis and according to Stockfish 14. But Stockfish also seems to think that I didn't make any serious mistakes in playing this line, which doesn't tally with my own analysis (as I have annotated above). Bb5 immediately wins 2 points of material (rook for bishop), but I had to find three separate forks in order to justify Ne5, and the first fork was almost fully refuted (I won a pawn). It seems like Black should have been able to avoid losing an exchange here with better play, but Stockfish says that Ne5 is only slightly less advantageous for White than Bb5.
Am I missing something, or is this just a case of "when one side is completely winning, the evaluation gets a bit imprecise" and in fact Ne5 was a poor move?