It's perhaps a bit like "everything is a model" in the sense that all of these tests, even the model-free ones, arise from a coherent choice of assumptions and,

It's perhaps a bit like "everything is a model" in the sense that all of these tests, even the model-free ones, arise from a coherent choice of assumptions and, if you for a moment take the Bayesian perspective very seriously, prior distributions over conditionals. The original quote should be taken to mean that any particular choice of assumptions is limiting, but making interesting choices can drive interesting questions which are thought provoking and meaningful even if they are wrong.