I would like to point out that the essential aspect of an "ellipsis analysis" of the NOC (or the null argument in general, or any other empty category, for that matter) is (well, should be) that it is represented "fully" at LF, containing all the structural information (and presumably the relevant lexical information as well) being analogous to a phonetically fully represented structure, whether such is a consequence of PF deletion, LF copying, or some other means.

As such, we should ask what definite predictions, in the form of a predicted schematic asymmetry in the terms of my CUP book, we can deduce under such an analysis (based on a universal hypothesis and a language-particular hypothesis, along with a bridging hypothesis in the terms of the CUP book) and how we can design Experiments to test such predictions. If we were to state this independently of the terms of my CUP book, we should ask what definite and testable predictions one can deduce under such an analysis, where the prediction-deduction necessarily involves a hypothesis about a universal property of the language faculty and one about a language-particular property, and how we can design Experiments to test such predictions.

That is how one would (try to) proceed if one were interested in deducing definite and testable predictions, as a means to accumulate knowledge of our subject matter.

If one were not interested in doing that, I have little to no idea how one would proceed. The answer seems to me to be sociological rather than scientific.