How do you approach cleaning and validating raw datasets?