The decisions that should exist before an AI feature reaches a public or internal launch decision.
Before launch, a product owner must define the user problem, the success measure, and the fallback plan if the model fails or slows down.
This checklist includes dataset quality, system logging, privacy review, and a response plan for edge cases and hallucinated outputs.
A feature without an owner is a demo. A feature with a named operator is a service.
