Purity (blue) and completeness (orange) for different parallaxbased CaMD cuts (along with bad parallax pool of giants) in the training sample. The choice of 50% (f=0.5) is justified by the plateau reached in purity and completeness, with a good balance between the two values. The discreteness of the plot is due to the small size of the training sample (N=53 666) used to test our cuts. All values are given in percentages.