A historic carriage-works helper invents a wheelwright who never worked there. Staff already have a vetted shop ledger. Which Domain 5 accuracy-security method should they require?
Select an answer to reveal the explanation.
Short Explanation
Think of an invented wheelwright who never worked there, and a vetted shop ledger on the shelf. Ground generations in retrieved approved text so the name can be checked. Temperature does not invent a ledger fact.
Full Explanation
The carriage-works helper inventing a wheelwright who never worked there is a hallucination accuracy problem with a security-relevant fix when staff already hold a vetted shop ledger. Require grounding generations in retrieved approved text from that ledger so every personnel claim can be checked against source passages. Retrieval-augmented generation with citation or quote constraints turns the ledger into the authority and reduces fluent fabrication.
Treating the need as a Domain 3 architecture lesson only fails by concept: knowing architecture without requiring ledger checks still ships invented names. Raising temperature fails because higher randomness increases—not corrects—fabricated identities; sampling does not create ledger facts. A dashboard title fails because charts are not shop ledgers and do not verify whether a wheelwright appears in approved records.
Exam caveat: grounding helps only if the ledger is current and the retrieval path is protected; a stale or poisoned corpus can still support wrong answers. Operational check: ask for the invented name with retrieval enabled, confirm the model refuses or qualifies when no ledger hit exists, and spot-check citations against the physical shop ledger before public release.