TAs and graders will submit data (e.g., scores) as spreadsheets. The top page will have the usual format, namely one row per student and one column per exam/assignment. Each exam/assignment will have a subsequent page with a column per question/problem. This use of spreadsheets minimizes the amount of deviation from current ways of doing things. Data can be entered into the spreadsheets automatically from the mark-recognition software, or it can be hand entered by graders and TAs.
Scanned documents will be stored as pdf bit images, and all other documents in Word or LaTeX format -- the .doc and .tex naming conventions prevent confusing the two. They could be converted to pdf, but we want to preserve editability.
From exams/assignments documents, we must somehow parse the questions/problems. Doing so requires a bit of help from the assessment's preparer. In LaTeX, the preparer can use an exam style-class, which is easily parsable and facilitates exam preparation. For those who prefer not to use LaTeX, a GUI-based exam-preparation tool is under development.
Once parsed, the individual questions/problems will be stored in item banks as XML documents or as special directories with local files for each attribute.