Experiment Track Record

Shows how well predictions held up. Ended experiments with a prediction are grouped by kind, fix type and metric, each group with its population, completion rate, hit rate (within or above the predicted range, weighted by evidence grade), median measured ÷ predicted improvement and inconclusive share. Stopped, unapplied and deleted experiments count as misses. Groups under 3 experiments return counts only: hit_rate and the median are null.