Why don't they try and develop a model that takes into account that students will study directly against the test?
The current model seems to rely on what could be called "security by obscurity". Of course students shouldn't be able to spend hours analysing a single text before the test, but if you publish a corpus of 200 texts, then anyone who read and analysed them thoroughly before the test aren't really cheating anymore, they are just educating themselves.
The current model seems to rely on what could be called "security by obscurity". Of course students shouldn't be able to spend hours analysing a single text before the test, but if you publish a corpus of 200 texts, then anyone who read and analysed them thoroughly before the test aren't really cheating anymore, they are just educating themselves.