That's a very negative attitude. Most of the catches and tricks we rely on are 'almost 100% fault proof' meaning that at certain levels one has to rely on statistics and probability. When we discuss checksums, CRC and all of that stuff, we sort of take it for granted that the checksum covers the contents .. always! So knowing that a built-in-self-test MAY fail, would that knowledge make the use of such tests unusable? I don't think so ! And, using the proper checksum calculation schemes, would it be logical (no, mr Spock!) that any single error would be overseen ? No, because that's why we invented these hard-to-code checksums in the first place.

So, in short, I DO believe (!) that self- tests ARE usefull. Like anything in life, they don't give absolute guarantees, but get you as close as possible to that under most circumstances.