Accuracy
Public accuracy reporting begins after at least 90 days of compatible archived evidence.
Why there are no scores yet
A small sample would look precise without being trustworthy. Forecast editions are archived from launch before metrics are published.
What will be measured
Temperature error, precipitation performance, wind error, forecast horizons, sample size, exclusions and model changes will be reported.
No retroactive rewriting
A provider change will create a methodology record and will not rewrite earlier forecast editions.