Re: Boo, hiss
The problem here is that the simple statistics of 2 out of 135 and 1 out of 19 launches doesn't tell you enough to make a true estimate of reliability.
You also have to include other events in shuttle launches, such as when they had an abort to orbit due to one engine shutting down early, which left them in the wrong orbit. Also the other cases where that almost had burn throughs on the shuttle o-rings on other launches, but didn't quite fail.
And I'm sure SpaceX has had other near misses as well, I just don't know of them for sure. But back tothe main point, you can't predict future reliability from such a small sample size, esp when you effectively throw the sucker away after each launch. This is why new airline designs do ground tests, then taxi tests, then high speed taxi tests, then first flight, then flight envelope expansion, then flight envelope testing, to make sure that the operating restricts are safely inside the "oh my god we're gonna die!" feelings, and then inside the "crap the wing broke and fell off" levels.
But testing one off items that get used and thrown away is much much harder. You need to have a process where you test as you build, test and measure, feedback into the production, track each lot of production, etc. It's a hard thing to do. This is what when you build artillery shells, you expect to have duds, misfires and other problems. And which you trace back to a particular lot, then look at the process, manufacturing, etc. And examine the un-fired ones for clues as well. But in that case... you're building thousands, tens of thousands or hundreds of thousands of shells and can do proper statistical analysis and prediction because you have so much data.
With rockets... not so much yet.