The counts below match the committed PostgreSQL 17 fixture exports. Load the fixture before rerunning the live demo; the generated portfolio dataset has different results.
- Model: Point to
findingsandfinding_eventsin the schema. A finding is one issue; several complete events describe its history. Explain why counting joined event rows would overstate findings. - Current picture: Run
make demo. Openartifacts/examples/fixture/backlog.csvandexceptions.csv. The fixture expects four actionable findings, one accepted risk, overdue F4, and missing-deadline F6. F10's deadline equals the cutoff, so it is not overdue. - History: Run the weekly report with end
2026-01-19T00:00:00Z; total backlog should be 4 → 3 → 5. Show F2's resolution and reopening. Three remediation episodes completed, although only two findings are resolved at the Jan 16 snapshot. False positives, not-applicable findings, and accepted risks are separate outcomes. - Rollout readiness: Open
rollout_progress.csv. The two nonempty rollouts should each be 50% complete. The empty rollout's completion is NULL. T2's later done event cannot affect the historical cutoff. - Evidence: Show the fixture manifest, test results, and both JSON benchmark plans once available. Explain one plan change and the late-arriving backdated-event limitation.
Try these on the dedicated test database or after a deliberate demo reset. Predict the result before rerunning a report.
- Join
findingstoassetsand state the grain. Then join allfinding_eventsand explain the extra rows. - Change the cutoff from Jan 16 to Jan 12. Which findings enter or leave backlog?
- Add a reassignment event after the cutoff. Show that Jan 16 ownership stays unchanged.
- Reopen a resolved finding after the cutoff. Explain how backlog and completed-episode throughput differ.
- Add an empty team. Check that R1 and R3 retain zero-filled output.
- Cancel a rollout task later. Compare eligible denominators before and after.
- Add a status-preserving
resolvedevent. Verify that R4 does not count another remediation.
Be able to explain LEFT JOIN, ROW_NUMBER(), LAG(), NULLIF(), < cutoff, and why p90 describes only completed episodes. The project is ready to present when these changes and the report outputs can be demonstrated independently.