A lot of data are naturally clustered for various reasons (particularly data with ids and timestamps). Also, note that BRIN indexes don't have to use ranges in

A lot of data are naturally clustered for various reasons (particularly data with ids and timestamps). Also, note that BRIN indexes don't have to use ranges in the summary--for instance, for enums or other domains of small cardinality, they could use bitmaps or bloom filters. I believe that BRIN indexes will become much more useful when those alternatives are implemented.