Pick a visualization tool you use (a weather app, a fitness app, a flight-search site, GitHub Insights, a stock app, a bike-share app, a university portal). Produce:
- Its data abstraction (items, attributes and types, dataset type).
- Three tasks it supports well, each as {action, target} on attributes.
- One task it supports badly, named precisely (which action and target fails, and why — what is not encoded or not linked).
- The chain a typical user follows (3–5 steps), marking where each step needs an overview, a selection or a comparison.
- One concrete change, and which level of the nested model, with which validation method, would test whether it helped.
Use a worked example as a guide: airline seat selection — items = seats; attributes: position (spatial), class (categorical), price (quantitative), availability (binary), legroom (ordered). "Find any free seat" = locate available items (well supported); "sit with my family" = locate adjacent free items (poorly supported); "is extra legroom worth it?" = compare two items on two attributes (badly supported: legroom is not encoded at all).