Tests can only detect failures, and can declare something as good only by elimination — i.e., when it did not find anything bad. Therefore, by definition, the PASS outcome cannot declare something GOOD that the FAIL outcome cannot declare as BAD
Before 11.6, we did not enforce this constraint, and that got modelers into trouble. For example, before you could model a TEST T as follows:
PASS outcome detection A and B, and
FAIL outcome detecting ONLY A
Well, not only is this contrary to the basic definition of a test — it also leads to some very illogical outcomes — B will be eliminated from the list of SUSPECTS, no matter what the outcome of the TEST T! Here is why — if Test T PASSes, you would declare A and B as GOOD, that is fine! However, if test T FAILS, then also you would declare B as GOOD (under single fault assumption) since A is faulty, and therefore B cannot be faulty! So, you do not have to see the outcome of TEST T, you can eliminate B!! So, why bother running TEST T, just eliminate B:-)
