One application query consumes far more RUs than expected. Which evidence is most useful before changing the schema?

1. Only the number of UI buttons
2. Query metrics, index utilization, examined versus returned data, and request charge
3. Only the application's display name
4. Only total repository size