….was it a fluke?

I think there are some category errors here:

In the vast majority of mechanical inventions, there have been thousands of trials at a component level, hundreds of partial (e.g. static tests of a rocket in which the engine is run but the rocket is not actually flown) or complete trials of a full system. It usually involves many attempts before a full system such as an atomic bomb actually works. Mechanical inventions that work right the first time are clearly the exception in the history of invention and discovery. Some possible exceptions are Tesla’s alternating current motor (if Tesla is to be believed), the atomic bomb, and the first flight of the Space Shuttle. Inventions that work right the first time do appear to occur, but they are rare, exceptions, outliers, flukes. They probably should not be treated as typical or likely for planning purposes or investment decisions.

…projects that succeed on essentially the first attempt are rare; in this, the Manhattan Project is quite unusual. Yet, this success of the Manhattan Project has greatly helped fund scientific R&D megaprojects that implicitly assume that the full system will work on the first try or with only a few attempts, something that is historically rare. Full scale systems like the ITER tokamak, particle accelerators like the Large Hadron Collider (LHC), and so forth are both extremely expensive and each trial of the full system is likely to cost anywhere from millions to billions of dollars. Thus, one hundred full system trials, perhaps a more realistic planning number, implies vast costs. Not surprisingly, many scientific megaprojects like the NASA Ares/Constellation program recently or the Super Conducting Supercollider (SSC) have foundered in a sea of rising costs.

Ignoring the fact that Constellation wasn’t a “science” project, one of these things is not like the other. Rocketry isn’t really rocket science any more. If you consider Falcon 1 a “training rocket” for SpaceX, consider that Falcon 9 worked almost without a hitch the first time (the only issue was the upper-stage roll), and Dragon worked the first time. If you do enough simulations, it is in fact possible to get it right the first time (though Shuttle had a pretty bad first flight — I’ve learned recently that Young and Crippen actuallywould have considered ejecting due to concern about the body flap damage from overpressure, had they known about it). The problem with vertical takeoff expendable rockets is that they pretty much have to work the first time, or at least in as few a number of tries as possible, because tests are expensive, and they’re not possible to incrementally test. I can’t emphasize enough what a breakthrough the new reusable suborbital vehicles are going to be, in their ability to incrementally test and do gradual envelope expansion. But in the context of incremental development and testing, I’m not sure what “work the first time” even means.

I would also note that Constellation’s problems were cooked in from the beginning, given what an awful design concept, and incompetent management it had. Combine that with the pork aspects, and its failure was inevitable, as many (including me) predicted at the time.

Anyway, despite the mixing of apples and oranges, it’s an interesting, albeit long, read.