A publisher's editorial team built a Q&A assistant grounded on their style guide and now wants an automated way, before wider rollout, to check whether the assistant's answers actually stay consistent with the retrieved style guide passages rather than drifting into unsupported claims. Which Microsoft Foundry capability addresses this directly?

The built-in evaluation metrics, including a groundedness score that measures how well responses align with retrieved source content, are the right choice because they directly measure whether the assistant's answers remain consistent with the retrieved style guide passages. That is exactly the drift check the editorial team needs before wider rollout: a response may sound fluent and still be unsupported, and the groundedness metric compares the answer against the source material it was supposed to use. The model catalog's cost comparison table option is wrong because it helps identify the cheapest model available and supports budget decisions, but it does not evaluate whether answers are factually consistent with retrieved content. The Content Safety hate and violence categories option is wrong because it screens for offensive or dangerous language, which is a different concern from ungrounded or invented answers; a response can be perfectly polite and still fail to align with the style guide. The increasing the number of tokens allowed in each response option is wrong because longer responses do not check accuracy or consistency with source content, and they can even give the model more room to introduce unsupported claims. The groundedness metric within Microsoft Foundry's evaluation tools is purpose-built for the consistency check the editorial team needs.