A proper scoring rule for evaluating probabilistic forecasts that measures distance between predicted and observed samples.