Yes! It could actually be both. Ownership over the recommendation system needs to include not just designing it and training it but also making sure that the up

Yes! It could actually be both. Ownership over the recommendation system needs to include not just designing it and training it but also making sure that the upstream data that feeds the model remains good. And I agree that a performance indicator is a potential solution, it just needs to be able to catch these kinds of problems.