Why is there a statistics difference in dataset details page profile and pipeline source stage profile?

In Catalyst QuickML, the profile on the dataset details page represents details for the whole data, but the profile in the pipeline source stage represents details for sampled data from the original dataset.