Yes, we can only describe with models what can be observed. But it is a bad excuse for ignoring contradictions in (or between) models.