You're evaluating an experiment where the outcome you care about happens very infrequently, such as a rare conversion or failure event in a product flow. The team wants to know how to reason about detectability when the primary metric is sparse.
How would you think about statistical power when the metric you care about is rare?