Let a constituent of a program by any of class, interface, method, field...

First step:

   1. Define a program p, composed of a set of constituent C.
   2. Define a set of refactorings R applying on any constituent.
   3. Define a set of metrics M to compute on constituents.
   4. Choose a subset of C, <c1,...,cn>.
   5. Choose a subset of R, <r1,...,rm>.
   6. Apply <r1,...,rm> on <c1,...,cn> to obtain <c'1,...,c'n>
   7. Compute the metrics in M on both the ci and the c'j.
   8. Apply the similarity measure and assess precision and recall.

Second step:

   1. Choose a set of programs P (either different version of a same program and different similar programs).
   2. Apply the similarity measure defined above.