You've shipped a model that looked strong in offline testing, but its production performance is now below target. The team wants you to figure out why the gap exists and what to check first.
How would you approach a situation where your model is underperforming in production?
