NavList:

Rafael Caruso, you wrote:
"Another one, perhaps more alarming, is that it was 'subject to mechanical failure. To recognize derangement, especially when the error is small, requires a thorough knowledge of celestial navigation.'  This remains true for all "black box" solutions to any problem still."

I'm not sure I agree with your last line here. Thinking aloud:

When the US Space Shuttles were designed in the 1970s, it became clear that they would have to be computer-operated at almost all stages of flight. But computers were not considered reliable c.1975. So what solution did they develop? How do we manage a deranged "black box"? They certainly didn't provide every shuttle astronaut with paper charts and tables so that they had the "knowledge" of astrodynamics to see through the black box. Instead they doubled down on the black box and allowed four independent computers to vote on the spacecraft's astrodynamics and flight operations (and they had a fifth backup computer running different software in case voting failed or "tied" but that backup was never needed in the history of the program). 

Consider the original "black box" in navigation, the chronometer. They infamously could become "deranged", and there was no indication of it. No navigator could open his or her chronometer and see evidence of its derangement, let alone correct it once it had gone bad. The early solution, and an independently useful method in any case, was lunars. The solution after about 1915 was radio time checks when radio was avilable. The normal solution after chronometers became common and before radio became inexpensive we know was similar to the shuttle's voting computers. The problem of the chronometer black box's lack of reliability was more black boxes! Carry three or more chronometers and treat them as votes on GMT.

Should Howard Hughes have carried three or more of these Maxson boxes in 1938? That probably would not have been practical with cost and weight coinstraints, and it seems that working a sight three times would erase any advantage from using the computer in the first place. Instead the redundancy might have been incorporated into the sight procedures. If you have a machine that can work the sights in half the time, then spend some of that saved time taking more sights. If the lines of position cross in a tight cluster, then you can be more confident that  the machine is working properly. This, of course, is a popular procedure today. Computation is cheap so we can detect errors and anomalies by taking more sights.

Frank Reed