Setting Examsis just like exposing photgraphs
Photographers know how important it is that pictures be properly exposed. If a picture is overexposed, much of the area will appear completely white. It is impossible, once the shutter has made the ever-so-cool sh-scht noise, to tell how bright the different parts of the picture are. All we know about the overexposed area is that they are all brighter than some value.
On the other hand, if the picture is underexposed, it becomes noisy. Instead of seeing varying shades of grey, splotches of colour will appear and there isn't a very good correlation between the actual colour and the colour on the photograph.
So how does this relate to exams? Exams are a method of determining how bright the students in a class are. An exam is--in theory, at any rate--a picture of how much the class has learned. In the same way that the quality of a picture can be compromised by over- or under-exposure, the quality of a set of examination results can be compromised by an exam that is took easy or too difficult.
If an exam is too easy, we can think of it as being overexposed. A large portion of the class will turn in a perfect paper and all we can tell about those students is that they knew enough to complete this exam perfectly. No further differentiation is possible without setting further exams.
This phenomenal is well understood by most university faculty. The generally accepted solution is simply to set an extrememly difficult exam. "You can always bell the marks up!". In photographic terms, they underexpose the exam. While this does prevent overexposure, it may lead to a noisy picture.
If an exam is extremely difficult, an element of randomness will be introduced into the resulting marks. Take three students, for instance, who are equally familiar with the course conent. Their marks may depend of which particular inane details they happened to have read while studying, or some other insignificant coincidence. If the average mark is 30%, it is very likely that some students who know more will receive lower marks than some of their less knowledgeable peers. Though the marks can be artificially "belled-up" to an average of 65%, the distribution will be very low quality.
This metaphor identifies two ways in which the quality of exam results can be compromised. The first, "overexposure", is well-known and understood. The second, "underexposure", seems to be have less of a presence in the general social consiousness. This is unfortunate, since the obvious solution to the first problem is likely to result in a manifestation of the second.