Teaching exam classes is theoretically straightforward. There should be motivated students, and a common goal, namely an exam to pass. Naturally, human nature intervenes and it's not so simple. For example, a class of Libyan middle managers would frequently complain to me "we're not just here to pass [the exam], we're here to learn English." Which was a reasonable observation, though you would need to qualify it if you knew that this class complained about everything from the perceived shortness of their prayer breaks, to the sugar in their tea.
But this got me thinking about possible research parameters around a real life testing and SLA situation. Say there's a commercial motive, in enabling students to pass a high stakes pen-and-paper test. One can of course provide classroom tuition for this, and no doubt would. But bearing in mind that most of this exam's candidates are teenagers, it would be useful in motivational terms to give them practice on an iPhone (or other) app. And it's in the design of the app where we get clever: make it a multiple choice with CAT, the students mail it into a program where it's scored with IRT. This would yield some very helpful data for that student, perhaps to be used back in the pen-and-paper classroom. But it would also yield warehouse loads of data if a large number of students were doing it.
Not only that, you could compare these quantitative results against the quantitative and qualitative data from the classroom AND from the results of the real world high stakes test, which is the rationale for all of this in the first place, to validate our app.