"test-to-failure" means that when you realize you may have made a mistake, you don't automatically scrap all your work to fix the mistake before continuing down the development path. You complete that prototype and test it because there are a thousand more mistakes to discover. Even if the known mistake will for sure result in a catastrophic failure, there may still be a large window of opportunity for productive mistake discovery before that for-sure failure ends that test run. You want to discover and fix mistakes in parallel, not serially.

Obviously there are judgment calls for when a mistake might prove too costly even in a test run. But manifestly SpaceX has damn good judgment in that regard. If you don't like how the sausage is made, just avert your eyes; don't complain about how ugly the sausage making process is.