Comment by ilaksh | Hacker News Reader

Comment by ilaksh | Hacker News Reader

When you make a change to a complex software/hardware system, you should expect something to go wrong the first time you test it.

Clearly they changed something for this test in particular and it broke the system. It is likely something fairly minor that just doesn't work when part of it is removed.

Also, if it just went through a retrofit and has actually never fired a missile, then it's not unlikely for something to have been not quite placed back where it was supposed to be.

This is the difference between doing a release with a QA process that takes some time and testing versus doing development and just launching the new version into production.

They changed the code or hardware for this particular test. That not working does not prove that the system has fundamental flaws or that other systems that have been tested and not modified will have the same issue.

Most people do not understand the inherent brittleness in engineered systems. Even those that are designed to be fault tolerant. That can cause confusion about outcomes and the necessity of testing every release.

It does depend on the approach, though. If you're SpaceX and switched on to this stuff, and testing is part of development that's one thing. If you've done your real testing already and this is more like a demonstration sort of test, that's different.

My theory is that they should have done a development test before the demo, but because it's so expensive , the higher ups just told the engineers to skip it.

They probably just said "Well, it was _only_ a retrofit, and you just changed a couple of minor things for the test. Are you sure we need to spend $17 million on another throwaway test?" And the engineer said he's not 100% sure it's necessary, but it would be advisable. So the higher up, not understanding how brittle and complex systems are or how development works, decided that meant it wasn't necessary.

I think the main issue might be that there is no way to do a good test without spending millions of dollars. Which I think might be because of structural issues in the defense industry.

Disclaimer: I'm no missile tech expert.

As far as I understand, these things have self-test routines and "dry-runs" which go through all the states in test mode and simulates a firing.

With my limited knowledge, this validates all the electronic systems on the missile or what it is, sometimes creating synthetic data during the test, so the missile can "navigate on the bench" by "wiggling it fins like a dreaming puppy".

Maybe this particular one passed that test, and something chemical failed along the way, I don't know.

That sounds extremely plausible. Maybe one of those tests had a || true still in its GitLab pipeline (-:

Or, more practically, they may be using pytest-vw (https://github.com/auchenberg/volkswagen).

About once a year I remember that package and laugh spontaneously.

> you should expect something to go wrong the first time you test it.

This was the second test-firing since 2016; that one failed too. I think that means they haven't done a successful test-firing in eight years. That doesn't inspire confidence (or awe, depending on who it's supposed to impress).

I don't understand your motivation to type this many words in order to make, what amounts to, generic excuses (1) for people that take a huge chunk of your paycheck in order to balance the world on the knives edge of nuclear deterrence and somehow managing to fuck up the one thing that is the whole point of the exercise.

Mind you, I am not even taking a moral stance here, I am just saying that if we have nuclear weapons can we please not lower the bar on competency in their handling.

1) maybe the reasons you give are plausible but they are essentially based on no real information about what happened here

I'm not making excuses I am trying to give my theory on their error which was simply to not spend the money to do a real test before doing a demo.

This was a real test. They have done this 192 times. It failed two times. The last two times. So it's not that this is inherently impossible, it's just that they can't do it anymore. I think, given the stakes, it's probably best not to just give them a pass.

Source for the number of tests: https://news.sky.com/story/trident-missile-misfired-and-cras...

Reply on news.ycombinator.com