Precision–recall curve of our neural network in both triage and vetting modes. The triage model is trained to distinguish PCs and EBs from obvious false positives, and the vetting model is trained to identify only PCs. The line labeled “vetting—plain” shows the original AstroNet model applied to vetting without the addition of any new features. The two dashed lines show the individual contributions of new features we added: “vetting—depth change” is the addition of transit depth differences alone, and “vetting—secondary eclipse” is the addition of secondary eclipse views. “Vetting—depth change + secondary eclipse” is the final AstroNet-Vetting model that combines both features.