Yes, I could find the bug quickly. It was not related to the individual formatting, but to the non-ASCII characters that you used in the rows > 20. In this case the SST contains a UTF16 string, but the UTF16-reading procedure did not check for CONTINUE records.