All sensitive attributes must be removed from the dataset

Each sensitive attribute value must appear an equal number of times in the dataset

The dataset should be completely synthetic

The distribution of a sensitive attribute in each equivalence class must be close to the overall dataset’s distribution