As the library should not expose its internal data model, it needs to provide its own statistical functions for reduction/aggregation.