maybe something wrong with the understanding of these puzzles and labels
As a chess player, I want to confirm one thing: the "label" in this dataset might not be the answer, but rather part of the puzzle. The true answer isn't provided here.
The style of puzzles in this Lichess dataset should be understood as follows: starting from a given position, after the first move is made, the opponent's correct response is what constitutes the puzzle. In other words, the first move is part of the interactive puzzle—a move that appears reasonable but actually contains a fatal flaw—and is not the answer itself.
Taking isobench/chess/puzzle/0m0X2 as an example, the actual puzzle is: after the white rook moves from b1 to b8, what should Black play? The answer to this puzzle is obviously not b1b8 (White's move), but rather Black's follow-up move c7c1, resulting in a checkmate victory for Black. If we interpret this dataset as treating b1b8 (White's move) as the initial move to solve, then this would be an extremely foolish decision for White.