Problem
A new model beats the old one on aggregate accuracy and ships. Two weeks later, support tickets spike: it's failing badly for users in one region on one device tier, a regression the aggregate number never had a chance of catching. Rollback takes 6 hours because nobody built one. Design the model promotion pipeline so a segment-level regression is caught before full rollout, and a bad one can be pulled back in minutes, not hours.
Required components
Click to add a component to the canvas.
Additional components
Browse by category. Click or drag a component onto the canvas.