In the case of partitioned datasets, the application layer is
responsible for data management, not float.