Calibration
Continuous verification: every nowcast is scored against the matching observation grid as soon as the truth lands. Aggregates are sample-weighted by cell count, so a verification covering 1M cells contributes 1M times to the means. Watch a real algorithm overtake the persistence baseline in real time.
Calibration · MAE / bias / RMSE
GET /v1/calibration · sample-weighted, grouped by algorithm × horizon
Loading…
Sample-weighted: a verification with N=1M cells contributes N times to the averages. The bar inside each MAE cell scales relative to the worst algorithm in the window — shorter bars mean a more accurate forecaster. The sparkline shows MAE per time bucket so you can watch a real algorithm's accuracy trend down over time.