I suppose it’s still a bad experience for the user if some % of attempts to connect fail or if some % of scripts/styles/images fail to load. So I think that means dns information about failures needs to somehow be propagated quickly. Not sure how well that works in practice.