Your question is Production Model Failure Recovery. 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).
Tell me about a time a model looked strong offline but failed after deployment. How did you diagnose what was going wrong, how did you work with others to stabilize it, and what did you change so the same issue wouldn’t happen again?