Observability Layer
Take the shortest observability exit first. The deeper contract section below only matters when the live-signal owner still is not obvious.
Observability data unavailable warning string, and the OpenObserve-first contract — and retired the landing's frozen July 10 framing: the July 10 refresh is now named as a point-in-time confirmation while the contract is taught as still-current. The monitor row's pointer still routes the cooldown-based alert path to the Workflows page, not on this district's streets.- Primary live source: OpenObserve remains the first telemetry destination; Watchtower stays the reader-facing interpretation surface.
- Route split: This district sorts the question first, then exits toward Watchtower, observability infrastructure, or troubleshooting.
- Best use: Open this page when you need the current explanation-versus-live-evidence boundary before following a deeper observability street.
- Source-health cue: If Watchtower says Observability data unavailable, treat it as an upstream OpenObserve load failure on that refresh, not a quiet event stream.
Choose the observability question you arrived with first
Keep these four observability layers separate
Treat OpenObserve as the place where current observability evidence lands first. The district should no longer teach a SQLite-backed collector as the main live model.
Watchtower README: raw event stream from OpenObserve. Stream: saffron.
Use Watchtower for live interpretation, alert review, and dashboard-level status questions. It reads and summarizes evidence; it does not replace the upstream systems that generated it.
Stay on the Watchtower street page when the live panel meaning is the blocker, not the infrastructure.
The service registry tells Watchtower whether a service is expected to be local, hosted, or deprecated, which prevents false “Down” readings for hosted services like OpenObserve on Railway.
Open the infrastructure street when you need the hosting and registry details behind a badge or health band.
The monitor loop remains the recurring health-check surface. Its monitoring.completed events still matter for alert cadence, service health history, and troubleshooting order.
Atlas documents the monitor run every 2 hours on the Watchtower guide and infrastructure streets; the full cooldown-based alert path is traced on the Workflows page in the Watchtower Alert Escalation workflow.
Use the district as a sorter, then leave
This district page is where Atlas keeps the boundary between explanation, evidence, and diagnosis explicit.
Street-level pages
These are the deeper exits once the current contract is clear and the district has done its sorting job.