Now, here is more about testing untestable systems.
Please read “He Who Can Not Be Tested” and “He Who Will Be Tested!” before you continue, if you want to be in the mood.

I found an article about this subject, which is called “Testing the Untestable: Reliability in the 21st Century“.

It is an article by the Los Alamos National Laboratory, and it has an ‘Approved for public release, distribution is unlimited‘ mark – nevertheless I am finding it hard to discover a link to the complete and final article, all I found is a comented draft here.
The complete article was supposed to be at this link (but the link is broken :(), and apparently it is available here at the IEEE Xplore for members. Probably you can ask for it at the IRRA (Institute for Reliability and Risk Analysis) page.

In the paper, they reinterpret the concept of testing and metrics, due to the difficulty of understanding testing of complex systems. They say, “constraints imposed by the real world inhibit the ability of researchers to calculate the reliability efficiently and accurately”.

In the article, between other things, they state that testing must also be built on subjective and intuitive criterias together with the accurate approach academy has to testing. I believe this is correct, and also think that the idea is not so innovative… just read the two first posts, and the thread in the newsgroup.

So… if you want to read the article, download the draft at this link.