What is the rationale of pair-wise versus three-way?
Most field faults occur due to low-level interaction between field values. Statistical design theorists have always believed that, and we have done our own experiments in which pair-wise coverage has routinely achieved code-coverage of 90-95%. The same kind of code coverage results have been shown by some independent clients. Further, we have measured the efficiency of the tests with respect to the number of field faults it found in “pre-tested” software — these numbers are extremely encouraging for the size of the testsuite it generates. Since it is clear pair-wise can never be exhaustive, and cutting costs is equally important, it becomes a matter of economics when one should go from lower to higher interactions. Of course, the algorithm achieves the same efficiency, but you inherently get more testcases. • Do you have an approximate measure of how the number of testcases changes when going from degree 2 to degree 3? The number of testcases grows polynomially with the interaction de
