A
You've shipped a model that looked acceptable during development, but it is now underperforming when used in practice. Your team wants a structured way to diagnose the issue and decide whether the problem is data, metrics, thresholding, or the model itself.
How would you approach a problem where your model is underperforming?
