This seems like it’s reversing the M->V generation. You don’t want to write your tests explicitly for the minutiae of the markup you are generating, but you don’t want to bypass the view layer entirely and go straight to the model because then you aren’t testing what you intend to. So you build a new model from the view layer, and then test the model. If the markup changes enough that tests start failing, then you update the model instead of all the failing tests.

I’m not certain that it buys enough to justify an extra layer of indirection in every front-end test though. Having a collection of selectors with meaningful names seems to get you about half the value without that extra layer.