Hacker News Reader

Comment by ForHackernews | Hacker News Reader

This is a good lesson in under-promise, over-deliver. NASA seem to be masters of that technique.

I saw first-hand how it comes to pass, from collaborating with NASA engineers years ago.

In order to ensure that a critical mission-goal is met, with, say, 95% confidence, every subcomponent must be much more reliable, as there will be many ways to fail. The net effect is that often the overall perceived reliability turns out to be much better than the requirement.

This is especially so for systems that have already achieved some measure of success, as a number of those subcomponents will have already done their jobs completely.

I always took it to be a budget defense method since they never get as much money as they as for, and have to expertly manage expectations to ensure they continue getting the “little” money they do.

Little is quoted because it’s a hotly debated and subjective word.

I suspect it's because of a solid physical engineering culture. They probably have enormous safety factors and redundancies or work-arounds, since you can't change anything physically once it's launched.

So if things go alright, those provide lifespan far beyond specifications.

> This is a good lesson in under-promise, over-deliver. NASA seem to be masters of that technique.

And Tesla seems to be master of the exact opposite technique with Autopilot.

It's far from clear that NASAs general conservative nature has been a good thing relative to a more risk-tolerant one. If we accepted more failure, could we accomplish a lot more?

Particular with SpaceX on the scene, it's time to start rethinking risk, because launch costs will be so much lower.

Have you seen the recent FSD videos? It's really quite impressive. It still makes occasional mistakes (that humans also make) but I find it quite astonishing how far it is.

Eh, at this point it's a boy crying wolf situation. Whenever there's a new space gizmo you know they're going to say "we only designed it to work for eight minutes", and then six months later it'll still be chugging along, because the actual design life was measured in years.

They could at least say "the doodad is designed to work for two years, but there's a 5% chance that it'll be DOA, because so many things can go wrong in space". That would be more realistic.

you don't deserve to be downvoted (this place is ridiculous) because saying that there's a "boy crying wolf" aspect is true, and it was the whole point of learning that story which point apparently some people missed.

but there is also a position I'd like to point out between the one most people here are taking ("NASA over-engineers every part individually to avoid failures")

and yours ("here's our reasonably expected life, but could come up short")

and that is: they need to set a bar that evaluates a reasonable amount of science they need to accomplish to justify the budget.

It's not the minimum number ("crashed in flames but we learned some things")

and it's not the gloriously optimistic number ("this thing's still going after 12 years, it's really worth it!")

and it's not a fake marketing expectations number ("we'll fake the budget numbers so we can razzle dazzle by blowing past the expectations")

rather it's just, "for X hundred million we expect to spend a month taking pics in every direction and gathering and analyzing some rocks from different promising places, and that mission alone is worth it" because that's what congress is approving

The problem is people hear what they want to hear. Much like telling a PM that a project is probably going to take 2 weeks but might extend to 2 months. Guess which number they are going to remember and hold you to?

Not really. They are putting experimental hardware into an environment that is not well understood. It is also being done by very public facing organizations with very large budgets. They basically have to guarantee that they will deliver what they say they will deliver. From a public relations standpoint, it may even be easier to justify a complete failure than an 80% success since a string of 80% successes will look like incompetence while a few complete failures are quickly forgotten.

It is always an exercise in horrible estimates and it gets really tiresome. If you always greatly exceed the estimate (by orders of magnitude!) it wasn’t a good estimate, it was PR. I assume NASA has real estimates they use internally. No one could be this bad at it and still have a job there surely.

Reply on news.ycombinator.com