Identical semantic models, but only one has an error

Hi JrBIAnalyst 

please try below points.
In your PBIX file, temporarily add back the removed column, even if it's just a dummy or blank column with the same name and type. Then, publish the file again using the exact dataset name to replace the old one. This should allow the Service to accept the publish without errors, as the schema will match what is expected. Once everything is stable, you can use Tabular Editor or the XMLA endpoint to safely remove the unused column from the dataset object, preventing issues with downstream reports.

If I misunderstand your needs or you still have problems on it, please feel free to let us know.  

Thanks.