Perhaps it requires more precise definitions, such as unique or primary key. All in all, evaluating aggregates with a quick read (fast-full-scan) seems acceptable to me. In current systems, with large read caches, it is fast. You also get the additional benefit of updating all the information in the table.