In page 126, "it is common to say that algorithm A is better than algorithm B if the upper bound of the 95 percent confidence interval for the error of algorithm A is less than the lower bound of the 95 percent confidence interval for the error of algorithm B .