That's my understanding: the math required to get a concrete prediction of a model is complicated enough that “either the theory, or the experiment, is wrong” excludes the most likely possibility: that too much error is being introduced by the approximations that they're (necessarily) using to get a tractable computation out of the model.
