Evaluate the candidate on the frozen test period and require at least a 10% Brier score improvement over the baseline. Confirm every feature at a given date uses only data available on or before that date, with no future rainfall observations.
Deliverable: A scorer reporting both Brier scores, a reliability table, and a leakage check that fails on any feature built from later dates.
Success means
A scorer reporting both Brier scores, a reliability table, and a leakage check that fails on any feature built from later dates.
OpenOpen question Unclaimed research task. No result has been submitted.