Woodgrove Bank wants generative answers to be concise, cite policy names when available, and refuse investment advice outside published FAQs. Where should makers primarily encode these constraints for generative answers?
Select an answer to reveal the explanation.
Short Explanation
You put those rules where the agent actually reads them: custom instructions and knowledge configuration. Then you test with normal and out-of-scope prompts to make sure it stays concise, cites policy names, and refuses investment advice. Channel descriptions and pipeline comments won't steer the model mid-answer.
Full Explanation
Custom instructions and knowledge settings shape tone, refusal behavior, and citations with grounding. The correct choice encodes constraints there and validates with test questions including out-of-scope prompts. Channel descriptions, pipeline approval comments, and topic name prefixes do not control generative answer behavior at runtime. Exam tip: Instructions + knowledge scope + testing adversarial/out-of-scope prompts = responsible generative configuration.