[esafak]

Doesn't that presume that one model dominates the other?

[nestorD]

It presumes some models are better than others (and we do find that providing data with a wide mix of model strengths improves convergence) but it does not need to be one model, and it does not even need to be transitive.