OMR was used by 8085 students in the first three weeks of term, including part-time students for the first time. It handed most students' registrations without problem, but there were a small number of problems that affected students. Of these, virtually all were due to departments not making up their minds what modules should be available (and with what cats scores and assessment methods) until well into term. This appears to be a deeply ingrained political problem, which never mattered much when all module registrations were captured on paper and could be sorted out over the following months. But now that there's an IT system to do this, the system needs to know what modules are running!
There were a small but irritating number of problems on the staff view side, caused by relational integrity problems in the database. These were due to a bug in a JDO superclass that I discovered late in September – the bug had been there for a year, but this was the first September that the system had been live and the very obscure set of circumstances when it did the wrong thing had never come up before. This was a sufficently fundamental part of the system that I had written unit tests for it, which reinforces my view that unit tests are not very good for testing correctness, especially written by the same person who wrote the code they're testing – they only test the object the way the author thinks it's going to be used. What unit tests are good for is regression testing: making sure you don't break something that used to work a particular way – even if that is not actually correct.