Rollout configuration

Read-only view of the wait-time Remote Config flags, resolved server-side. Values cannot be edited here.

Current wait times

* Canonical confidence — read from the same waitTimes documents the apps consume, not from a shadow or snapshot collection. A lane with a wait but no confidence value shows Not available, never zero.

Preview shadow — legacy vs robust

Isolated replay of the robust estimator against real reports. These numbers are not published and no app reads them — robust is what the wait would be if the robust estimator were canonical.

Recent comparisons

Summary by lane

* Confidence shown is the published canonical value for context. The preview does not recompute confidence and does not change it. A difference of 0 min (agree) means the two estimators produced the same minute — it is a result, not missing data.

Canonical pipeline activity

Data activity only — this panel shows when wait-time documents last changed. It is not function execution health.

Active crossing sessions

Aggregate counts from activeWaitPresence. No user, session, or location data is read or displayed.

Overall health

Pipeline

Steps marked monitoring not yet available have no backing Firestore aggregate. Function health is observable only in logs — a document existing is not evidence a function is healthy.

Model accuracy

* Median / P90 are approximate, interpolated from absolute-error histogram buckets. MAE, bias and all rates are exact. Models with no scored samples show Not available, never zero.

Evaluation health

Unmatched reasons

Shadow & snapshot health

Lane breakdown

Monitoring not yet available

These panels need a backend aggregate that does not exist yet. They are listed rather than estimated, because producing them today would require scanning collections the dashboard must never scan.