AI-103: Developing AI Apps and Agents on Azure

Groundedness evaluation metrics in Microsoft Foundry observability measure whether a generated response is actually supported by the content retrieved for the agent, so a drop in groundedness after a content update points directly at the retrieved content as the source of degraded answer quality. Indexer status and failure history only indicate whether ingestion succeeded, latency breakdown traces show performance timing, and prediction drift metrics track changes in model output distribution over time — none of these directly reveal whether the retrieved content is undermining response accuracy.