@AdamDavis gave some really good low-level finds in the parent comment. Has that been passed on to the developers?
Sounds like he found the symptom on the reading-in-of-the-data. Now we just have to find what is writing the incorrect FieldName out in the first place, eh?