I think this bundles together several importantly different things. First, “not currently testable” cannot simply mean “untestable”. Deutsch has proposed that quantum computation could in principle test distinctly multiversal consequences: the explanatory claim is not insulated. More importantly, Everett was not invented as imaginary evidence to rescue quantum theory from falsification. Almost the reverse: the ad-hoc move is collapse. Everrett noticed that we have universal unitary quantum dynamics and then, when “measurement” occurs, Bohr stipulated a special transformation - collapse - without explaining physically what constitutes measurement or what performs that transformation. Everett says: don’t add that exception; take the dynamics seriously everywhere. On the wizard criterion, that matters enormously. Deutsch’s 2016 paper “The Logic of Experimental Tests” explains why the multiverse is a better explanation than collapse even though “everything possible happens”; it also develops the explanation centred view I am here defending as a refinement of Popperian epistemology, which was originally focused more on propositions which only approximate scientific explanations. That said, Popperian falsification was never “one anomalous observation → throw the theory away”. Popper explicitly recognised auxiliary assumptions, experimental error, problematic observations and competing explanations. What he opposed was conventionalist stratagems: modifications whose function is merely to immunise a theory from criticism. So “special administrative measures” are perfectly compatible with Popper provided the theory remains exposed to criticism rather than being protected from it indefinitely. And I think a contrast with your Bayesianism actually reveals the deeper disagreement between us. You ask how we could “objectively” distinguish the alternatives if we cannot assign meaningful base rates/likelihoods. Exactly! I don’t think epistemology consists in assigning probabilities to rival world-models in the first place. There is no given probability distribution over possible explanations from which rationality tells us which one to buy. Bayesian updating tells you how to redistribute probabilities conditional on a model space and likelihoods you already supplied. It does not explain where those explanatory world-models came from, why these are the relevant possibilities, or why one explanatory structure is better than another, nor can it factor in explanations that we haven’t thought of yet. Conjecture and criticism do that work. That is also why I resist the sociological conclusion. Of course textbooks, careers, equipment and institutions affect which theories people investigate. Kuhn taught us something important about scientific communities. But it doesn’t follow that explanatory preference is constituted by convenience, usefulness or social commitment. Otherwise we struggle to explain scientific progress itself: why enormously inconvenient conceptual revolutions sometimes succeed despite entrenched institutions. The relevant distinction isn’t: testable now = science untestable now = ad hoc/religion. It is closer to: Does the conjecture solve the problem by introducing independently constrained explanatory structure, with consequences and counterfactual reach beyond the anomaly it was introduced to address? Or does it merely stipulate whatever is required to save the theory? The latter is ad hoc. The former can be a genuine explanatory conjecture even before we possess the technology to perform its decisive experiment. Occam doesn’t rescue the alternative. Everett is the nice example precisely because “one universe” sounds ontologically cheaper, but achieving it requires adding the unexplained collapse machinery. Counting entities is not the same thing as counting arbitrary assumptions. A vast world generated by a simple, independently constrained explanation can be less ad hoc than a small world requiring a wizard.

Deutsch’s “wizard” criticism reformulated: An explanationless theory stipulates an overall transformation while supplying no explanatory account of the subsidiary transformations whose composition would make it possible.

Not true (see Karl Popper’s The Myth of the Framework”). However to have a meaningful conversation both parties need to want to improve their prior understanding of the situation