[torpfactory]

It kind-of depends on what sort of issue you're finding in your test. Complex systems like spacecraft often have unexpected interaction effects which only testing can reveal. I would call these the 'good' kind of test learnings - ones that an army of great engineers wouldn't be able to predict.

This is editorializing but it looks like Boeing didn't uncover very complicated interactions - they failed at a more basic level of competency - timer synchronization for the launch issue and then a major software bug for an important orbital maneuver. Those sorts of issues really should be sorted out on the ground using hardware simulators. Furthermore, the timer synchronization failure prevented testing of the docking hardware, further delaying the overall program.

For a human rated vehicle, personally I think you should have at least one full-up, fully nominal test before you send anyone along for the ride.