An insurer's claim agent cites a policy rider that was withdrawn last quarter. That withdrawn file is still in the grounding index, beside the current rider. What should the architect correct first?
Select an answer to reveal the explanation.
Short Explanation
A binder that still holds the rider the company withdrew will keep getting quoted. You remove that withdrawn file from the grounding index so answers land on the current rider. Training a model on the old rider copies the mistake into the model.
Full Explanation
Grounding timeliness means the agent can see the current source and cannot still retrieve a source the business has withdrawn. The claim agent cites the withdrawn rider because that file is still indexed, while the current rider is already beside it, so the first correction is to remove the withdrawn file. Fine-tuning a model on the withdrawn rider teaches the retired text as if it were still the policy, which makes the stale answer harder to retire. Adding trigger phrases for the word rider changes how a topic is selected and does not take the withdrawn file out of the index the agent is already reading. Replacing the claim agent with a small language model that memorizes the withdrawn rider freezes the same retired wording and drops the current rider on the floor. Removing one stale file is enough here because the current rider is already indexed. Exam caveat: timeliness is about which revision is in the grounding set, not about how firmly the prompt says to be current. Operational check: search the index for the rider, confirm the withdrawn file is gone, and ask the same claim question again.