[Feedback] :pencil2: correction — Scope cloud retention D1/D3/D7/D14 by signup cohort week

## Feedback (correction)

**Question:** Scope cloud retention D1/D3/D7/D14 by signup cohort week
**Tool:** `query`
**Dataset:** `semantic.scope_cloud_events`

### Notes
Initial Scope retention query incorrectly used dateDiff(first_event, last_event) as a proxy for retention — this measures lifetime span, not actual return events, and inflates early cohorts. Correct pattern: use min(event_date) AS signup_date in a subquery, then LEFT JOIN back to scope_cloud_events AS e, and use maxIf(1, dateDiff('day', signup_date, e.event_date) = N) in SELECT. Remember: inequality in JOIN ON not allowed — put date filter in WHERE clause. Also: scope_cloud_events only covers cloud inference sessions; local Scope usage is entirely untracked (no telemetry dataset exists for local mode).

---
_Feedback ID: `da34cfc7-97c6-470b-bb49-66238d804a50` | Recorded at 2026-03-09T21:54:21.545Z_

Start with the query tool and the semantic.scope_cloud_events dataset described in the issue. Replace the lifetime-span calculation with signup_date from a min(event_date) subquery, join back to the events, and validate D1/D3/D7/D14 using the stated WHERE date filter; confirm that the result is limited to cloud inference sessions.

Written by the indexing model from the issue text.