Netflix homepage experiment with conflicting metrics

Primary disagreeing with secondaries (CI crosses zero, p=0.14) means we cannot rule out noise on the headline metric. Title Starts +0.90% is real; Completion Rate -0.60% signals more clicks but worse per-click satisfaction - a "clickbait" pattern. Recommend an extended ramp or a follow-up experiment which keeps the algorithm but injects a long-term-retention surrogate (30/60/90-day return rate, churn-risk model uplift), so the ship decision is tied to satisfaction, not clicks. Don't ship on Title Starts alone: the experiment is incomplete. State the decision explicitly: "extend 4 weeks with a retention-secondary; if Completion returns to baseline, ship; if not, abandon." Add a heterogeneous effects slice for TV (largest share) to ensure UI fits.