As they say, hindsight is 20/20. It might be relatively easy post-hoc to trace a problem that has already occurred back to a relatively small set of causes, an

As they say, hindsight is 20/20. It might be relatively easy post-hoc to trace a problem that has already occurred back to a relatively small set of causes, and then ask why these were overlooked - after all, an ounce of prevention is worth a pound of cure. But when the system itself is so complex that any of a thousand other factors might have gone wrong instead, you're really comparing a thousand ounces of prevention to a pound of cure. You can get really caught up in all of the plausible contingencies, which grow disproportionately with complexity. If you were to audit that oil rig inch-by-inch, you might have found dozens of things which could have led to a problem of near the same impact as the current leak. And you might have even focused on fixing all of them. But you can't presume that some other factor that you didn't notice (or couldn't have noticed given the finite nature of your knowledge) wouldn't have triggered a slightly different failure, and you can spend an enormous amount of time fixing things that would not have caused a failure. Adams makes a good point - the best way to ensure the reliability of a system is to understand the functioning of the system as a whole, and beyond a certain scale of complexity, this is extremely difficult and often impossible.