A pathology system posts bloodwork results into Salesforce with a status of "Pending Vet Review" before a vet signs off, then "Released" afterward. The group never wants the agent to disclose a pending result to an owner, even if the model reasons the owner is asking urgently. What is the most reliable configuration to enforce this?
Select an answer to reveal the explanation.
Short Explanation
If a result is still "Pending Vet Review," the safest design doesn't let the agent hold it at all. A filter that only ever fetches "Released" rows means there is nothing pending for the model to slip up and share, urgent-sounding request or not.
Full Explanation
The reliable enforcement mechanism is a filter on the query behind the lab-results action that restricts eligible records to "Released" status, so a pending result is never fetched into the conversation in the first place; there is no language for the model to misjudge because the data never arrives. A classification description only decides which topic handles the utterance, so tuning it against urgent phrasing does nothing once the lab-results topic is already correctly selected and an action runs. An instruction telling the agent to "be cautious" is guardrail-style guidance and is meaningfully better than nothing, but it is applied at the language layer and depends on the model correctly interpreting an urgent, emotionally charged message every time, which is exactly the reliability gap a filter closes. A longer scope statement describes the topic's purpose in general terms and has no mechanism for preventing a specific record from being returned; scope communicates intent to the reasoning engine, it does not gate data. The exam pattern to remember: when "never" has to be absolute, reach for a filter, not a longer paragraph of instructions.