BRIN is great for huge tables where data is naturally ordered on disk (e.g., time-series by created_at). It’s tiny and fast to maintain, but less precise than B-tree and relies on good correlation with physical order.