If you were given a dataset with obvious discrepancies right before a major presentation, how would you handle it?