We conclude with a series of positions where a potential bishop fork looks unplayable because the forking square is guarded—but where the fork nevertheless is the winning move because checkmate or some other decisive consequence results if the forking piece is taken. We have seen that a parallel logic can make seemingly bad knight forks quite productive; naturally the same is true of any other sort of fork that may not look feasible at first.
On the left is a first example of the idea. Look for a visual pattern; see that White has his king, rook, and d4 pawn arranged in a classic triangle that calls for consideration of a queen or (here) bishop fork by Black. (You could also just ask what checks Black’s bishop has; the only answer is Bxd4.) Bxd4+ thus has the potential to take the rook at a1, but the d4 pawn is protected by White’s knight. There's no good way to get rid of the knight, but don't stop there. Imagine the sequence failing and ask what it would make possible. It then goes 1. …Bxd4+, 2. NxB. What lines would then be open? What checks would Black then have? The answer is Qe1—mate. So the bishop fork at d4 works after all, winning a rook and a pawn.