Evidence based on Internal Structure
Traditionally: construct validity, discriminant validity, convergent validity, nomological validity.
What are the relations between items or groups of items within your test? What should they be? Should you measure one construct or two? Three?