> Anecdotal story: I had a bug a few weeks ago that all sorts of people thought was really dumb / obvious / shouldn't have happened. I checked... the code hadn't been changed for FIVE YEARS. Nearly every customer we had ran this code for five years. Maybe I feel that they should have had a catch for the situation that occurred, but whomever wrote it wrote code that lasted bug free for five years ... maybe they weren't dumb and they knew something I didn't?

I've seen this. Huge big O complexity but on a dataset that fits in cache? Runs fine. As soon as the dataset gets bigger, you get systematic timeouts.