I just read "The Challenger Launch Decision" (after someone here recommended it) and it has a more nuanced view of what went wrong. While there was a long history of concern about the O-rings, there was also a history of getting conflicting data, of implementing "fixes" that appeared to work, only to come back later. Even the engineers themselves, while expressing concern like the memo mentioned in the article, they did not, until the launch decision conference the night before, voice anything as strong as "stop the launch".

During reading the book, I was struck by two things: one was the "slippery slope" introduced by first seeing something and arguing it was not a launch risk. With the decision making system in place, that meant it was very difficult to then, when another datum came in, to say that the two data together implied something was wrong, because they had already argued that the first datum was not a risk. This is of course a well-known phenomenon, but it didn't seem like the system was well-equipped to deal with sparse and sometimes contradictory data.

The second was the feeling that what was happening was that the engineers were observing a random process and post-rationalizing cause and effect. (I had just read "Fooled by Randomness", which may have contributed to this.) Every piece of information was made to fit into some model, but it seemed like no one was considering that maybe what they were seeing was just inherent randomness. With that view, the progressively bad outcomes that happened before indicated that what was observed was a poorly-characterized random process with a fat tail. At some point, one rationalization was that they had a "safety factor of 4 left", but if you have indications of a fat-tailed process, that's not much to bank on.