Currently we write all timestamps as full timestamps (with seconds). At a minimum it would be better if we determine if time is not included in the original data and then not include it in the synthetic data. Better still if we detect the datetime format and replicate it in the output.