Slopes and mean values of linear regressions computed on overlapping four-hour windows of mean lifetimes retrieved from the survival model, used for the training of support vector machines. We plot a set of decision contours in data space given (arbitrary) time-decision boundaries as targets for the SVMs. As the time boundary increases, the contours deform continuously, demonstrating self-consistency with the shift of boundary. The nonflaring (NF, gray markers) class separation remains similar for every time boundary. The scores for the different boundaries are shown in Fig. 12.