AI-300 decision sheets

• Azure AI Search retrieval: hybrid vs vector vs full-text vs semantic ranking
• Azure ML automation auth: OIDC federated credential vs service principal secret
• Foundry built-in roles: User vs Account Owner vs Project Manager vs Owner
• Foundry deployment options: standard vs serverless vs managed compute vs provisioned vs batch
• Foundry and Azure OpenAI monitoring: App Insights vs diagnostic setting vs signals
• Foundry resource scope vs project scope
• General-purpose evaluators: relevance vs coherence vs fluency
• Sweep sampling: grid vs random vs bayesian and early termination
• Managed VNet isolation: allow_internet_outbound vs allow_only_approved_outbound vs disabled
• MLflow on Azure ML: tracking URI, run status and where to register
• SFT vs DPO vs RFT vs Distillation vs Retrieval
• Model monitoring signals and their reference datasets
• Endpoint type: managed online vs Kubernetes online vs batch
• Private endpoint DNS: why the hostname still resolves public
• Prompt techniques: few-shot vs priming vs grounding vs chain-of-thought vs clear syntax
• RAG evaluators: groundedness vs relevance vs retrieval vs F1 vs response completeness
• Responsible AI dashboard: which component answers which question
• Azure ML workspace dependencies: which are required, which is optional