If a dataset includes the descriptions of two families and 4 genera but doesn't tell you which genera belong to which family you will be forced to get this information from someplace else (e.g. ITIS). If the dataset is based on a different arrangement from ITIS you don't know this because the dataset didn't tell you because it's not part of the standard. In this case the data won't make sense because the family descriptions need to be a superset of the genera which belong to them. I think this is a pretty basic problem if we want to support hierarchical data (taxa at different taxonomic ranks).
Again, if you don't want to follow the author of the dataset then you are free to ignore the suggested classification - but I would strongly suggest that you better know what the author is thinking and ignore all of her data if you don't agree with it, not accept the descriptions while rejecting the classification.
Steve