Run n times, compare results. The idea of the same bit flipping twice is fairly ludicrous. These n times could be on the same machine, or distributed and compared with paxos. Then back everything up a billion times. Safety is extremely hard to guarantee, so the closest approximations will be expensive.