Slide through the year after launch as a model's accuracy slips from 94% with no error or warning. The reading turns red once it falls below 85%.
Nothing broke. Nobody changed the code. The model simply gets slowly worse — and nothing in the system announces it.
