You can produce all the videos you want extolling the virtue of learning from experimental failure, but blowing the whole thing up, repeatedly, is a lot worse than normal experimental failure. They don't have the design of this thing adequately analyzed if they are this many iterations in and still can't keep it in one piece. Major faults that would destroy the whole system should be found in the component testing phase - which is I'd guess is where they are short cutting. It can be expensive creating good subsystem test rigs.