Your question is Model Underperformance Debugging. 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).
What is your approach to debugging a model that is underperforming in production?
Explain how you would determine whether the issue comes from data quality, distribution shift, training-serving skew, model logic, infrastructure, or an incorrect evaluation process. Describe the measurements, comparisons, rollback criteria, and validation steps you would use before deploying a fix.