How would you test whether a new synthetic data approach improves rare-event performance for Autonomous Solutions?
