1. How should an interaction pattern be chosen for an AI feature?
2. Which pattern sits at the highest-trust end of the ladder?
3. Why is the draft pattern the most consistently successful? Select all that apply.
4. What is described as the highest-leverage design move in an AI product?
5. Why is banded or behavioural confidence preferred over a numeric percentage?
6. What are the three parts of a good refusal?
7. Why must refusal behaviour appear in the evaluation set?
8. A team deletes the pre-AI rules-based classifier at launch because the model performs better. What has this done?
9. Why does labelling every AI-touched surface damage a product?
10. What relationship should hold between the empty state and the evaluation set?