You download the data and complete your analysis with ample time to spare. Then, just before deadline, your collaborator lets you know that they've "fixed a data error". Now, you have to do your analysis all over again. This is the reproducibility horror story.