Run arrival measured
When each forecast run went live on this site, read from the publish workflows' own records: the moment the live pointer was written, and how long after the model's cycle time that was.
These are measurements, not a service level: Xue makes no availability commitment.
Reading metrics…
BY SOURCE
Latency runs from the cycle time to the moment the run went live, so it includes the model's own production time. All times are UTC.
Each strip is the window: one bar per cycle that went live, at its cycle time, its height the latency against the source's own longest; a mark under the line is a cycle that never arrived.
* Typical upstream: an experience value for when the model's own data is usually complete after its cycle time, not a measurement. Blank where there is none.
AT THE DEADLINE
For each local day, the newest run already live at the cutoff, and how old its cycle was then.
LANGUAGE