It's both a PR/optics thing, as well as a real engineering thing.
Say you have 5 critical components, any of which fail, you are no longer able to complete your entire mission (which in this context is a set of scientific experiments that you've pitched and believe that if you get nothing but that data back, that your entire mission cost was worth it). If you need X time to finish your mission, and say... you want a 95% (I have no idea what the real model number is) chance of finishing the entire mission. This will require you to push the reliability of your individual components up a lot. Say your failure rate model is uniform - your probability of failing in any given period of time is the same as any other given period of time (so you're in the bottom of the bathtub curve).
Very roughly you'd need like a maximum of 1% chance of any given component failing over X period to reach your reliability goal. If X is something like month, that translates into a mean time to failure that is not suddenly measured in years. Very quickly you can see how the constraint that N=1 MUST survive for X period will naturally lead to many instances where that N=1 CAN survive for much longer.
Now say that you can withstand component failures, and still continue with degraded performance (that would have impeded your original mission), then you can stretch things out even more.
Say you have 5 critical components, any of which fail, you are no longer able to complete your entire mission (which in this context is a set of scientific experiments that you've pitched and believe that if you get nothing but that data back, that your entire mission cost was worth it). If you need X time to finish your mission, and say... you want a 95% (I have no idea what the real model number is) chance of finishing the entire mission. This will require you to push the reliability of your individual components up a lot. Say your failure rate model is uniform - your probability of failing in any given period of time is the same as any other given period of time (so you're in the bottom of the bathtub curve).
Very roughly you'd need like a maximum of 1% chance of any given component failing over X period to reach your reliability goal. If X is something like month, that translates into a mean time to failure that is not suddenly measured in years. Very quickly you can see how the constraint that N=1 MUST survive for X period will naturally lead to many instances where that N=1 CAN survive for much longer.
Now say that you can withstand component failures, and still continue with degraded performance (that would have impeded your original mission), then you can stretch things out even more.