Following the autor definition: " The first agent, a 1 , operates an Auto-correlation feature evaluator, @ 1 . The evaluates the worth of a subset of features by determining its pitch period. A features considered effective if its correlation coefficient scores less than 0.95, otherwise, it is omitted."
Then the first agent needs first to filter the features with more than 0.95 auto-correlation coefficient, and then determine the worth by the pitch period.
An efficient way to test if the first agent performed well enough is to compare with the table 2 of the paper.