Catalog models in Microsoft Foundry already score most fraud types for an insurer, and they miss one rare label. The agent should call a model for that label. What should the design specify?
Select an answer to reveal the explanation.
Short Explanation
You hire a specialist for the one illness the clinic keeps missing, and you check that specialist before the first patient. Design a custom Foundry model only for the fraud label the catalog misses, and wire it to the agent only after an evaluation set passes. Rebuilding every label is a different project.
Full Explanation
A custom model is justified for the behavior prompting and the catalog models do not produce. The catalog already scores most fraud types and misses one rare label, so the design is a Microsoft Foundry custom model scoped to that label, and the agent must not call it until an evaluation set has shown it meets the bar. A custom model for every fraud label, including the types the catalog models already score well, throws away working catalog behavior and widens a training job the gap does not require. The catalog model alone, with the rare missed label accepted as something the agent will not see, leaves the stated gap in place and calls it a design. A prompt action with no evaluation set, asked to catch the rare label the catalog models miss, skips both the custom model the gap calls for and the evaluation that must precede any wiring. Exam caveat: the evaluation set is a promotion gate for this label, not a sample the agent may consult at runtime as if it were knowledge. Operational check: confirm the custom model's label list is only the missed label, and confirm the agent has no tool binding to that model until the evaluation set has been scored.