Failure modes under out-of-spec conditions have to be identified and characterized. You have to be able to predict what will happen when the device is "abused" in the sense that its environment/power supply/etc. are outside the specified limits. It's difficult and costly, and, in fact, difficult to identify which effects are due to which failure modes. It would be "nice" if, whenever such limitations are exceeded, the device simply stopped, but that's seldom what happens.

Even more troublesome is the task of identifying which effects are primary and which are secondary, i.e. caused by primary effects. If properly characterized, some of these can be localized by their occurrence in commercial-range testing, but generally, it's a crapshoot, and it takes some pretty clever engineering to isolate faults that are caused by other faults.

RE