An explainability score for end-to-end driving

End-to-end planners produce a trajectory and no reason for it. Define a measurable score for how well a model’s stated reason matches what it actually did, and evaluate it on BEVDriver.