Whatever the system is designed to protect against should be used to attack it (with dummy warheads, of course). If the system is designed to stop, say 20 simultaneous ICBMs, then we need to fire 20 ICBMs at it simultaneously. Or whatever. But piecemeal testing will NEVER prove the system, no matter how many times it is done.

Let me give an example. I was principal designer for a the new safety warning system used in USA's gasseous diffusion enrichment plants. After a couple of years, and upteen thousand lines of code, it was tested and accepted. It worked 100% fine. Right? Wrong!

The tests, which were thorough and exhaustive took place from Monday through Friday, week after week, until the system was approved. The idea was that, if the system could work, error-free for a week, it would work forever. Nonsense.

It turns out that the system crashed after about two weeks on a 24*7 basis. This kept happening. It was finally traced to a failure of the compiler to deallocate memory. This required a new compiler, and still more testing.

NOTHING less than a TOTALLY REALISTIC FULL-SCALE test will prove a damned thing. The system might work OK for 7 targets, and fail totally with 8. Who knows? Nobody! That is why FULL-SCALE testing is absolutely necessary.