In this case, proving a negative about unexpected acceleration of Toyota cars:
Sudden acceleration in Toyota vehicles not an electronic issue, U.S. study finds
I had thought of calling this post a Car Talk question for comp.arch.
When I got my car back from the dealer from servicing the last time, it did something weird and a little scary, like suddenly decide that it didn't want to go anywhere right after I had pulled out of a parking spot. The service manager offered to examine the car again, but I told him I'd rather see how difficult it was to reproduce the problem.
The problem occurred just twice and hasn't happened again in three months. My suspicion is that the "servicing" put my stateful vehicle into some weird state that was not easy to duplicate, and that, having gone twice through some critical section of code, the car has forgotten all about the experience.
The only people I know of who do enough testing with machines that are stateful in ways that are hard even to enumerate would be people who work with microprocessors. I wonder if anyone at a water cooler at NASA has noted that it wasn't Intel, but someone else, who discovered the problem with the Sandy Bridge chipset.
Now, I'm a bit of a cranky old carmudgeon, and my instinct is to shrug my shoulders at my car dealer and at NASA and say, "Oh, well, I never thought you knew what you were doing, anyway." Intel doesn't know what it's doing either, but by now it either knows it or should know it. NASA has proven beyond a reasonable shadow of a doubt that it is uneducable.
My question is this: does anyone in this business believe that it is possible to uncover all corner cases by any plausible testing protocol that someone would actually spend the money for? By "all corner cases," I mean enough that corner cases would never be discovered in the wild after having escaped discovery in the testing labs.
Robert.