Reading here. It looks like many properties are missing. Parameters to the constructor are documented. But most of them are then copied to properties of the object. But the fact that those properties (and which, and when (when just metadata is downloaded or when also data is download) is available on the object is not really documented. I see just id and openml_url.

So it is also unclear how stable it is to access those properties, and which ones are optional and which ones are not. For example, If i call openml.datasets.get_dataset in Python API, will I always get back object with md5_checksum property set or not? What if I call openml.datasets.get_dataset with download_data=False?