Your question is Design a Multi-Armed Bandit. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Explain how you would design and evaluate a multi-armed bandit algorithm for content recommendation at Netflix.
Discuss how you would represent user, title, device, and session context; choose an exploration strategy; use logged recommendation data; and avoid bias from the existing policy. Describe offline evaluation and how you would validate the approach through a controlled Netflix experiment. Include the main reward definition, safety constraints, and production considerations.