Some datasets have more nominal values listed than they really use

At least for the following datasets:

openml_dataset_152
openml_dataset_153
openml_dataset_186
openml_dataset_49
openml_dataset_51

I have found that they have more nominal values listed than they really use. This is especially a problem on target column, because it changes multi-class classification task to binary classification task.

For example dataset 51 has the following nominal values listed for its num column:

    "<50",
    ">50_1",
    ">50_2",
    ">50_3",
    ">50_4"

But if you check here you see that only 2 unique values are present for that column. Why are other nominal values even listed? Could you cleanup this?

Also, why this API call contains the number of unique values in the column, while this API call does not. It would not be necessary to list if the list of nominal values would be correct, but it is not.

Compare the nominal values for datasets 152, 153, 186, 49, and 51 between the linked features API and dataset pages, starting with the two API responses for dataset 51. Trace where these metadata values and unique-value counts are produced; done means unused nominal values are handled correctly and the two API representations are consistent.