Thanks for the response. We've been looking into it a bit more of course, and found a bunch of entries in the upgrade logs that refer to
"Ignoring extra finish states in attempt at question 295"
"Two inconsistent finish states found for question session 7447 in attempt at question 309 keeping the later one."
I'll send over the results of that query shortly. In the meantime, does this issue seem like something can be solved, or is it more serious and, er, catastrophic than that? Mainly I just want to determine if we should roll back to 2.2 ASAP, to minimize data lost if we can't fix this, or if it seems like there is a reasonable chance of a fix, we can save the hassle of a roll-back and keep working on it...