Description
Rank 2 - high value, moderate cost. Live usage-insights lines carry datasourceType and panelPluginId but generic scenes merges multiple apps. Add a bounded per-stack view of datasource-type and panel-type query mix through the existing exact instance_id guard, without claiming per-app attribution. Proposed new emitted series: 0; point-in-time detail is a view.
Acceptance Criteria
- #1 Top-N view is bounded and scoped to exact instance_id on every query
- #2 Dashboard explains that datasource type is inference, not app identity
- #3 Raw datasource or plugin values never become metric labels
Definition of Done
- #1 just test
- #2 just tf-validate
- #3 just check-identifiers and just no-em-dashes both return clean
Implementation Notes
Parked at Wave 1 boundary: researched build candidate in doc-0006, outside this wave implementation scope. Resume when Rob selects the next product work; first verify the candidate metric or datasource contract named in this task.
Wave 2 code landed at a070885: bounded top-20 plus remainder query mix per live stack, scoped by exact instance_id, no new metrics. Four implementation attempts; final CodeRabbit had one valid non-finite input finding, reproduced and fixed by root. Focused 154 passed, 794 subtests. Guarded dev read found 368 and 183 requests on active stacks with datasourceType, but no non-empty panelPluginId; panel positive proof remains unavailable. Final exact-SHA gate, CI and dev readback pending.
Final Summary
Code acceptance: exact instance_id query guard, bounded top-20 and remainder per stack, inference banner, zero new metrics or raw-value labels; Gate G and hosted CI 35935370824 passed at product 51ce670. Live dev datasource-type source evidence exists, but the new S3 view was not published because T2 failed before scan and dev was rolled back. Positive panelPluginId proof remains unavailable.