We would need to know the distribution of incorrect answers. If it didn't give any other answer more than twice then you could trust the 13-time answer to be the correct output of the computation.