No, the samll test dataset has only 32MB. I created the dataset by using Dataset.create(...) datasset.add_files(...) and then dataset.finalize() . I unfortunately dont have s3. I poked around in the saved data on the share and it seems that for some reaseon folders 'data' to 'data_11' have their contents deleted. Whats even weirder is that they were deleted right at the time when i first tried to get a mutable copy today, the other folders are untouched since monday when i created the dataset. I will remake the dataset again, but any ideas why this happened?