There is still some work to be done to complete our analyis. It must be studied how the current set of typing rules needs to be extended to cover features like external functions, constraints, partial application, and search operators. This should not be difficult but has just not been considered yet.
The form of the effect might be changed to reflect some more exact information about the program points that raise the nondeterminism.
The major task is the development of a type inference algorithm for our type and effect system. The user is not forced to specify any type by hand (as in mode based languages like Mercury), but of course every type given by the user might increase the precisenes of the analysis.
If the analysis is not able to prove the deterministic behavior of some part of a program (which is actually deterministic), the user will have the possibility to specify information about functions thats deterministic behaviour she can guarantee. Thus, we will allow the user to specifiy types which are always considered correct by the analysis, even if the analysis cannot prove this.
Last but not least, we will consider to refine the type domain by
using a three element domain 
where denotes
expressions that evaluate to head normal form. This should increase
the precision of the analysis by some amount.