Before releasing a RAG-based support-chat model, a team runs Azure AI Foundry's built-in evaluation tools to score, without a human reviewing every response by hand, how well each generated answer stays faithful to the retrieved source documents and how directly it addresses the user's original question. (Select TWO.) Which two evaluators would give the team these two signals?
Select all correct answers, then click Submit.
Short Explanation
The team wants two different questions answered automatically about every response: did the answer stick to what the source documents actually say, and did it actually address what was asked. One evaluator checks the first thing by comparing the answer's claims against the retrieved material, and a separate evaluator checks the second thing by scoring how directly the response matches the user's question. Neither of those jobs has anything to do with money or network speed. A budget alert only tells you when you are about to overspend, which is a completely different concern from whether the chatbot is making things up. And how fast the storage account responds tells you about plumbing speed, not about whether the content that came out the other end was faithful or on-topic. The two evaluators that actually measure faithfulness to sources and directness of the answer are the ones built for exactly that purpose.
Full Explanation
The correct answers are C and D. Groundedness scores whether the claims made in a generated answer are actually supported by the retrieved source documents, directly measuring faithfulness to the source material, while relevance scores how directly the answer addresses what the user actually asked, together giving the team both signals they need without manual review of every response. Option A is incorrect because budget alerts from Azure Cost Management track spending against a threshold; they say nothing about whether an answer's content is accurate or on-topic. Option B is incorrect because storage account network latency is an infrastructure performance metric describing how fast documents are retrieved, not whether the retrieved content was used faithfully or whether the resulting answer addresses the question. Both distractors describe operational or financial monitoring rather than content-quality evaluation, so neither can substitute for the groundedness and relevance evaluators that Foundry provides specifically to automate answer-quality scoring in a RAG pipeline.