The target fitness function is clever. Basically, it is "symmetric difference cardinality".

That is, the closer the bits in the hash result for the candidate are to the target hash result, the better it considers the candidate (in particular, cardinality of symmetric difference for a bitset is the count of the number of one bits in the symmetric difference. It then sets the fitness to mean lower is better, so the smaller the number of different bits, the better it considers the candidate)

This means it just tweak input key bits through mutation until the result comes out right. Now, in theory, there should be no correlation, so this should be no better than random search, but ...