Below are the notes sent to michigan and the ones returnes:

From Columbia:
http://www.columbia.edu/cu/libraries/inside/projects/apis/system/dataloads.html

From Michigan:
1.      Structural Metadata formatting

        + hopefully it is correct this time
        
2.      Multiple "associated name" parsing (many)

        + should be vastly improved, though probably not perfect

5.      Titles with Shelfmarks or other non-title information as content  
(ca. 740), e.g.,

        + the majority of these were addressed by omitting the records from  
the submission, which is OK because they were only record placeholders  
really.

6.      HTML character entities for special characters (> 1700)

        + most have been filtered


The following were not addressed. I'm presently thinking that these  
need to be done manually.

3.      Citations (dd510s) with corrupted data (ca. 15), e.g.,

4.      DDBDP Citations with Problem Data

7.      Unknown" for authors (ca. 1300)

8.      Duplicative date information in dd245_a, e.g.,