In reverse-engineering a double coin flip differential model, what would be the amount of falsified "yes/no" responses in your dataset?