From memory (documentary on previous similar tests and FEA modelling a long time ago), testing achieves the following:
- shows the design and construction matches model predictions during normal conditions (say up to working loads)
- shows the same up to design loads (often working plus a percentage e.g. 150% of cabin differential)
- shoes the reliability of the model at the extreme edges (testing to failure)
When testing to failure, you often find out if the nonlinearities in real life were adequately built into the model (think of all the various stresses causing strains and deflections in different directions combining with each other) and - hopefully - you have correctly predicted the weakest point and failure mode at which structure eventually fails.