[Daishiman]

He's saying that things that have such a low occurrence that they're indistinguishable from statistical outliers are not good targets for generalized public policy, since the benefits are not provable.