I think the missing value is treated as a text variable in the case of fill-in-the-blank items. It may be that it is not converted to numeric in cases where the user changes the fill in the blank item to numeric data. In such cases, I would imagine SPSS would convert the text data to 0 (as all text equals 0 in the numeric world) but that the CSV and TXT files would contain the original (pre-converted) missing values (e.g., the 99's in this case). I'll add this to our to fix list for the next version.