Probabilistic scores are now on the leaderboard

The boards now show SQL (Scaled Quantile Loss) next to ELO and MASE. It is the mean pinball loss over the nine deciles, scaled by the same naive forecast MAE as MASE, so the two read on one scale. Lower is better. Definition: fev-bench.

This is a first step, and only models that submit quantile forecasts are scored on it. A model that submits a point forecast alone shows a dash instead of a number. ELO is unchanged and still based on MASE.

If your model submits quantiles and you see a dash, open an issue at github.com/DAG-UPB/ts-arena and we are happy to help.