Adam Witt

The goal is to learn how lightweight you can make each component before it becomes unreliable. Every gram of weight you save is another gram of payload you can deliver. Theory and lab testing can only get you so far, so you have to try making components lighter & thinner and see if they fail in flight. The simplest optimization cycle is to shave a little weight off each component, launch, and see if something fails. When nothing fails, you make everything a bit lighter and try again. When something fails, you make that component stronger and try again. I forget the derivation, but under some very simplistic assumptions the procedure to improve the fastest gets you 1/e success rate. A real project is more complicated. You can often get partial failure information, like after a mission success you can see a component is slightly damaged. And different missions have different stresses. If you launch successful on a day when the winds were very calm, you wouldn't take that as a signal that some components are too strong. Also, failures cost more than successes. But overall, the optimal failure rate during R&D is not close to zero.