I show how to use metadata to avoid costly reads when retrieving counts, even for a subset of the table.