Your question is Production Drift and Quality Evaluation. 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).
How would you design a robust evaluation pipeline for monitoring production model drift and output quality?
Discuss how you would collect production data, define drift and quality signals, handle delayed or sparse labels, compare results across segments, and determine when to alert, retrain, roll back, or require human review. Address both batch analysis and near-real-time monitoring, including training-serving skew, privacy, cost, and false-positive control.