add the extra rule that if there is only one possible instance the compiler should use it.

Yes, I think that is what I am saying. The closed class assumption alters the types of exported functions by applying improvements to them - that is not what I meant at all. I think what is confusing me is the 'possible' part. I was thinking about applying type improvement to effectively determine which instance to choose. (I think this is an error on my part - I get the impression type improvement cannot be used for this) ... would a possibel instance be something like: assuming you are resolving instances lazily at the last possible moment, you try to unify all instances and if one and only one unifies - use that instance. Keean.