A financial services compliance team is rolling out a Foundry-hosted assistant that answers employee questions about internal trading policies. Auditors require that every answer include a reference back to the specific policy document and section it came from, so a reviewer can verify the response against the source. Which approach should the team configure in Foundry to meet this requirement?
Select an answer to reveal the explanation.
Short Explanation
Auditors here aren't just asking is the answer right, they're asking can you show me where that came from, and that's a different requirement than accuracy alone. The way to satisfy it is to have the assistant reach into the actual policy documents each time a question comes in, pull the relevant passage, build its answer from that passage, and hand back a pointer to exactly where it looked, the same way a well-footnoted report works. Baking the policy text into the model through retraining might make the answers sound right, but it erases the paper trail a reviewer needs, and it quietly goes stale the moment a policy changes without any built-in way to flag that. Letting the model write longer answers doesn't create a reference back to a source, and making its wording more varied run to run actually works against consistency, which is not what an auditor wants. What auditors need is traceability, and that comes from retrieving and citing real documents, not from a longer or more creative response.
Full Explanation
The correct answer is D. Retrieval-augmented grounding connects the assistant to the actual policy documents, retrieves the relevant passages for each question, and generates the answer from that retrieved content while attaching citations back to the source document and section, which is exactly what the auditors need to verify each response. Option A is incorrect because a larger output token limit only allows longer responses; it has no mechanism for tracking which document a fact came from and does not add citations. Option B is incorrect because fine-tuning bakes patterns from the training documents into the model's weights, producing answers that sound like the policies but carry no traceable link back to a specific document or section, and the underlying policy text can also become outdated without a way to flag it, which is the opposite of an auditable, verifiable answer. Option C is incorrect because temperature controls how much variation appears in wording across generations; increasing it makes answers less consistent and does nothing to connect a response to its source, which works against the auditors' verification requirement. Grounding with retrieval and citations is the feature built specifically for traceable, source-backed answers.