The self-refutation charge is too slick. A methodological rule (“prefer scientifically testable beliefs”) isn’t a proposition that needs to pass its own test, any more than “drive on the right” is false because it isn’t itself a car.

This is the strongest form of the reply. A methodological rule still needs a justification, and the justification (this method is reliable) is a substantive claim about the world that is not itself a scientific finding. One can adopt the rule pragmatically, but then one has conceded that at least one important belief, the rule’s own reliability, is held on non-scientific grounds, which is the point. The rule survives as a preference; it does not survive as a limit on what can be known.