How would you design an A/B test to evaluate a new video recommendation algorithm at Tubi, and what primary and guardrail metrics would you track?
