I have been working on code for analyzing the moves in a game of Fischer Random Chess to calculate what the original position was.
It depends entirely on the move notation you start from. In PGN game records castling would be written as O-O or O-O-O, so you would always know whether a move is a castling. Even in other notations for FRC castling is usually not written as the King step, because this would cause ambiguity when the King only moves a single step. In the UCI protocol for communicating with chess engines castlings are therefore encoded as the King 'capturing' the friendly Rook it wants to castle with. That would also make in unambiguously a castling, when there were no moves from or to the destination square before.
I don't think the problem can be solved in general. The first move of a Knight or Bishop could come from two locations on the back rank. And if both these locations are visited by an enemy piece before the occupant of the other moved (or the game finished before the other moved), you would never know which of the two it was.
It depends entirely on the move notation you start from. In PGN game records castling would be written as O-O or O-O-O, so you would always know whether a move is a castling. Even in other notations for FRC castling is usually not written as the King step, because this would cause ambiguity when the King only moves a single step. In the UCI protocol for communicating with chess engines castlings are therefore encoded as the King 'capturing' the friendly Rook it wants to castle with. That would also make in unambiguously a castling, when there were no moves from or to the destination square before.
I don't think the problem can be solved in general. The first move of a Knight or Bishop could come from two locations on the back rank. And if both these locations are visited by an enemy piece before the occupant of the other moved (or the game finished before the other moved), you would never know which of the two it was.