There are lots of ways of generating (pseudo-)random numbers. Is it sensible to conceive of a protocol for this, allowing bindings of *random-state* to objects of different classes to affect the behaviour of random? We probably need to finish the work on subclassable structures for this to make sense.