Rather than grading a free-form essay, we ask the model to pick one of three machine-readable labels that capture how much information it claims to have: limited, intermediate, or extensive. Using discrete categories keeps the output tidy and directly comparable. Template 1 offers only these three options, forcing the model to rank its expertise. Templates 2 and 3 add a fourth escape hatch: NA, for when the model wants to confess ignorance. Template 2 presents this four-way choice as a one-line prompt, while Template 3 adds brief definitions of what limited, intermediate, and extensive should cover. Comparing answers across the trio shows whether the model's willingness to hallucinate drops when we give it an explicit path to refuse and when we clarify the scoring criteria.