Consider ck.has_no_nans(df_modeling_data); where I only want to cause an error if my dataframe contains NaNs before it goes into scikit-learn.

It would be informative to add an error message that is reported to the user, something like ck.has_no_nans(df_modeling_data, message="scikit-learn requires no NaNs to be present"); so we have some context along with our check.

Might it be sensible to add optional messages to the checks? The benefit would be the accumulation of documentation about the expected output of transformations and the expected state of ingested data which otherwise might be forgotten or written in comments which may not keep up with the checks themselves.