You didn't do anything wrong. Sphinx needs to do that to traverse the imports to build the API section. We're actually in the process of speersting kedro datasets into its own package so this will be cleaner in the future