We speculate that the Mockitator tree’s poor LPO success rate was due to Mockitator’s already high coverage (and correspondingly high ZeroR success rate). A larger number of projects would likely improve LPO crossvalidation, since loss of one project would have a smaller impact on the success rate. LPO cross-validation also indicated how well a decision tree would predict coverage on individual projects. Figure 5 shows the success rate on each project when left out of the training set, relative to each tool’s ZeroR success rate. “N/A” indicates that Randoop crashed while generating tests. We found that the success rates for the Agitator (Figure 5(a)) and Randoop (Figure 5(c)) trees were higher than ZeroR’s success rate on most projects.