The median anomaly score for each class computed for latent representations of transients obtained from full light curves when a single isolation forest is used for anomaly detection [bottom] and when MCIF is used [top] (this is the exact same as Fig. 5, reproduced for convenience). The scores are derived from the unseen anomalous data and the common transient testing data. The five classes on the right (scores in bold) are anomalous. The common classes have somewhat lower median scores when using a single isolation forest, but the common classes SLSN-I and AGN (among others) are considered very anomalous, unlike when using MCIF.