(Left) The confusion matrix for the best-performing single-epoch triplet model, as calculated using a test set of 2400 candidates. The confusion matrix is shown as a one-vs-all binary classification for reals, though the distribution of examples are balanced across classes. The percentages shown represent the fraction over the total number of sources for a particular class. (Right) The ROC curve for reals, characterized as a one-vs-all binary classification. We determine a threshold of 0.6 for true positives of reals at an FPR of 1.94%. The inset shows a zoom-in of the top-left edge of the curve, used to define the score threshold.