Approvals & the analyst loop

Every experiment needs a person's approval. Nothing reaches shoppers automatically, and no winner is rolled out automatically.

Every 15 minutes the analyst:

1. evaluates running tests (and pauses any with a traffic-split problem or clear harm);
2. checks data quality, and pauses recommendations if events look unhealthy;
3. refreshes evidence-ranked opportunities;
4. drafts the best next test for a free page type into Needs approval.

You review the exact copy, placement, metrics, risks and rollback plan, then approve. When a test finds a likely winner, you decide whether to roll it out.