A generative feature has been live for six months. Usage is steady, no incidents have been raised, and quality has never been measured since launch. Which stage of the lifecycle is missing, and what is the consequence?

• AFeedback, and there is no route from a poor answer back into improving the system.
• BDeployment, and the feature is running without a supported hosting arrangement.
• CModel selection, and the team may be using an unsuitable model.
• DData selection, and the training corpus may be biased.