Observability — UI surfaces¶
Target
This page describes intended behaviour or structure. No implementation is claimed unless a section is explicitly marked Current implementation or Evidence.
The Observability platform serves two Studio work areas: Operations & usage (timelines, overviews, status, objectives, alerts) and Quality & releases (evaluation, evidence, feedback, proposals). Every screen reads a projection or an evaluation record through the Studio BFF (GW-02); no screen reads a backend directly.
Screens¶
| Screen | Work area | Reads | Acts | Roles |
|---|---|---|---|---|
| Run list | Operations & usage | ListRunProjections + Operations.Subscribe |
open run, filter by state and cost status | Requester, Reviewer, Operator, Observer |
| Run timeline | Operations & usage | GetRunTimeline; artifacts and receipts via their owners |
drill to trace (backend link), open approval, open receipt | Requester, Reviewer, Operator, Observer |
| Tenant and project overview | Operations & usage | GetOverview, GetSubjectOverview |
change window; drill to agent, tool or model | Workspace admin, Operator, Observer |
| Status and objectives | Operations & usage | ListServiceHealth, ListSlos |
tighten an objective (admin) | Operator, Workspace admin, Platform admin, Observer |
| Alerts | Operations & usage | ListAlerts, ListAlertRules |
acknowledge, resolve with reason, edit tenant rules | Operator, Workspace admin, Platform admin |
| Evaluation runs | Quality & releases | ListEvaluationSuites, GetEvaluation, ListReleaseEvaluations |
start an evaluation, request a waiver (approval) | Asset maintainer, Reviewer, Platform admin, Observer |
| Release evidence | Quality & releases | GetVerdict + evidence artifact |
compare target vs baseline per metric | Asset maintainer, Reviewer, Observer |
| Feedback | Quality & releases | ListFeedback |
record rating, correction, comment, escalation | Requester, Reviewer, Operator |
| Datasets | Quality & releases | ListDatasets |
snapshot a dataset | Asset maintainer, Platform admin |
| Learning proposals | Quality & releases | ListLearningProposals |
pick up, accept (opens a Draft release), reject | Asset maintainer |
States the screens must show¶
- Run per ADR-0014 vocabulary; usage per FUX-08 (estimated, reserved, observed, reconciled, uncertain shown as separate figures, never one total).
- Projection freshness: every operations screen shows projection lag; a stale projection is labelled, not hidden.
- Objective: Meeting, BurningFast, Breached, Recovered, Suspended with error budget remaining.
- Alert: Fired, Acknowledged, Resolved, Suppressed with occurrence count.
- Evaluation: Queued, Running, Scoring, Completed (verdict Passed, Failed, Waived, Inconclusive), Failed, Cancelled; a waiver shows its approval and reason.
- Proposal: Proposed, InReview, Accepted (with the resulting release), Rejected, Superseded.
Other clients¶
- CLI:
factory runs show <id> --timeline,factory slo list,factory alerts ack <id>,factory eval run --suite <id> --release <id> [--baseline <id>],factory eval verdict <releaseId>,factory feedback add. - VS Code extension: run timeline panel for the active run; inline feedback on a generated file; evaluation verdict badge on an asset.