Selective scientific realism (SSR) answers the pessimistic meta-induction by recommending commitment only to the 'working posits' of successful theories. Boge (2021) sharpens the pessimistic induction by arguing that in at least two research fields — nuclear structure physics (liquid-drop vs. shell models) and psychometrics (Rasch vs. two-/three-parameter item response models) — it is exactly the mutually strictly incompatible posits that are essential to the empirical success, so SSR cannot deliver a general explanation of scientific success. The paper leaves open whether this challenge can be met: whether SSR can supply principled, non-ad-hoc criteria for which weakened posits deserve realist commitment without collapsing into empiricism or immunization; whether strict incompatibility of essential posits is exceptional or a general pattern, and whether the existing cases dissolve into unifying limiting accounts; how the probability-skewing (the γ\gamma terms) needed for a Bayesian vindication could be grounded in actual theoretical practice; and, if SSR cannot be vindicated, what alternative framework could systematically explain the prevalence of empirical success.

Background

The debate over scientific realism asks whether successful scientific theories are at least approximately true. The 'no miracles argument' (NMA) holds that the empirical success of science would be miraculous unless successful theories were approximately true; the 'pessimistic meta-induction' (PMI) replies that historically, many empirically successful theories (phlogiston theory, caloric, the ether) turned out to be false, so present theories are likely false too. The standard realist rejoinder is selective scientific realism (SSR), in its deployment- or working-posits variants: only the posits that were actually responsible for a theory's empirical success are claimed to be true and preserved in successors — for example, relativity theory extends the inertial mass mm to the velocity-dependent expression m/1−v2/c2m/\sqrt{1-v^2/c^2}, thereby abandoning the generic frame-independence of mass while still recovering the Newtonian predictions at all speeds small relative to that of light, so it is arguably only the weaker posit actually doing the work that survives theory change.

Dawid and Hartmann (2018) gave a Bayesian reconstruction of the NMA on which a frequency estimate of the success rate in a research field, combined with a weak truth–success connection, yields a lower bound above 1/2 on the probability that a successful theory is true. Boge (2020) attacked this reconstruction with an incompatibilist datum: suppose the successful theories in a field fall into kk mutually strictly incompatible classes (theories within a class are compatible or related by limits; theories across classes contradict each other in a way that no limiting relation removes). Since at most one class can contain the approximately true theories, an objective Bayesian should bound the probability of truth by p(t∣c)≤1/kp(t|c)\le 1/k — where the condition cc denotes this incompatibilist datum, i.e. that the field's theories fall into k≥2k\ge 2 compatibility classes — which for k≥2k\ge 2 conflicts with the Dawid–Hartmann lower bound. 'Strict incompatibility' (Boge 2021, Definition 4) means that two theories contradict each other with respect to their essential posits; for mathematically formulated theories, the paper's LC criterion judges strict incompatibility to obtain when the theories make contradicting claims and neither is the limit of the other, where a 'limit between theories' is made precise in Definition 5: T1T_1 is a limit of T2T_2 if for each successful empirical prediction of T1T_1 there is a mathematical limiting procedure by which T2T_2 recovers the prediction.

Boge (2021) turns this into a direct challenge for SSR. He defines a success-continuum of theories (SCT) — a time-indexed family of theories in one field, partially ordered by recapture of empirical successes — and calls a posit 'relatively essential' to an SCT if it figures non-trivially in the derivation of empirically successful predictions and is not replaceable by weaker posits across the succession. The SSRist is entitled to be a realist only about the surviving, relatively essential posits. Boge then argues, via two case studies, that it is sometimes exactly the relatively essential posits that are strictly incompatible across rival theory families: in nuclear physics, the successful liquid-drop and shell-model families — rival models of the atomic nucleus whose essential assumptions disagree on whether nucleons move by and large freely, as shell models require, or are tightly bound, as liquid-drop models require (a disagreement about the mean free path of nucleons inside the nucleus) — have no unifying view available (he assumes 2≤k≤42\le k\le 4 compatibility classes); in psychometrics, the theory of psychological measurement, the Rasch measurement model and the two-/three-parameter item response theory (IRT) models — rival families of models for scoring test-takers' item responses in terms of latent traits — achieve predictive successes while disagreeing essentially about the parameters governing item response, with no success-preserving limiting relation between them (k≥2k\ge 2). A numerical study of the resulting Bayesian expression fk,γ(e,g)f_{k,\gamma}(e,g), where γ∈R+\gamma\in\mathbb{R}^+ parametrizes how much 'skewing' information could raise the probability of truth above the equivocal 1/k1/k, shows that only special circumstances — an anticipated unification — would allow the realist to constrain fk,γf_{k,\gamma} significantly.

The paper is candid about what remains open. On delimitation, it states that the general conditions under which a posit does 'real' work in deriving a successful prediction are 'an open question', and it warns that the realist's recurring retreat to ever weaker posits 'threatens to collapse the realist position into an empiricist one, or at least verges on an immunization strategy'. On evidence and scope, it concedes that the challenge stands only 'so long as such a dissolution is missing' for the case studies, and it does not establish how widespread strict incompatibility of essential posits is across the sciences. On probability, it leaves open how the required γ\gamma-skewing could be non-ad-hocly grounded in theoretical practice. And it allows that if SSR cannot be vindicated, it may become more attractive to explain scientific success in some alternative way — though the paper can only gesture at possibilities, such as a variant of a Darwinian account of the prevalence of success or an account of misleading evidence.

Problem Statement

Can selective scientific realism be vindicated against the incompatibility-based pessimistic induction, or does the challenge stand? Concretely, the open questions posed by Boge (2021) are: (i) the delimitation question — what principled, non-trivial criteria can SSR supply for which (successively weakened) posits legitimately count as the real targets of selective commitment, such that the weakening has a non-ad-hoc endpoint and does not collapse into empiricism or an immunization strategy; (ii) the scope and evidence question — are the known instances of strict incompatibility among relatively essential posits (nuclear physics: liquid-drop vs. shell models, 2≤k≤42\le k\le 4; psychometrics: Rasch vs. 2P/3P item response models, k≥2k\ge 2) exceptional, or indicators of a general pattern across the sciences, and can detailed domain-specific evidence — mathematical limiting relations in the sense of the paper's Definition 5, derivational analyses, or independent tests of confounding variables — decisively establish either strict incompatibility or a unifying limiting account in each purported case; (iii) the probabilistic question — how, if at all, could the prior-skewing information (the γ\gamma terms) required to keep a Bayesian truth–success connection viable in the presence of k≥2k\ge 2 compatibility classes be grounded in concrete theoretical practice without ad hoc assumptions, given that on Boge's analysis only anticipated unification would suffice; and (iv) the fallback question — if SSR cannot be generally vindicated, what alternative explanatory framework (e.g., a Darwinian account of the prevalence of successful theories, or an account of misleading evidence) can systematically explain the prevalence of empirical success. These components form one coherent objective: determining the epistemic status of selective realist commitment in the face of mutually incompatible yet successful essential posits. The four components are not a mandatory checklist: the central vindication question and each of (i)–(iv) can be addressed independently, and together they form the single objective stated above.

References

  1. Boge, Florian J. (2021). Incompatibility and the pessimistic induction: a challenge for selective realism. European Journal for Philosophy of Science 11(2):46. DOI 10.1007/s13194-021-00367-4. https://doi.org/10.1007/s13194-021-00367-4
  2. Boge, Florian J. (2020). An argument against global no miracles arguments. Synthese 197(10):4341–4363. DOI 10.1007/s11229-018-01925-9. https://doi.org/10.1007/s11229-018-01925-9
  3. Dawid, Richard & Hartmann, Stephan (2018). The no miracles argument without the base rate fallacy. Synthese 195(9):4063–4079. DOI 10.1007/s11229-017-1408-x. https://doi.org/10.1007/s11229-017-1408-x
  4. Boge, Florian J. (2023). The Positive Argument Against Scientific Realism. Journal for General Philosophy of Science 54(4):535–566. DOI 10.1007/s10838-022-09636-z. https://doi.org/10.1007/s10838-022-09636-z
  5. Golemon, Luke & Graber, Abraham (2023). A deductive variation on the no miracles argument. Synthese 201(3):81. DOI 10.1007/s11229-023-04078-6. https://doi.org/10.1007/s11229-023-04078-6
  6. Nyström, Johannes (2025). How to estimate the success chance of a scientific theory? On the no miracles argument and the base rate fallacy. European Journal for Philosophy of Science 15(4):63. DOI 10.1007/s13194-025-00699-5. https://doi.org/10.1007/s13194-025-00699-5
  7. Boge, Florian J. (2025). Rethinking holism and underdetermination. Synthese 206(4):192. DOI 10.1007/s11229-025-05278-y. https://doi.org/10.1007/s11229-025-05278-y
  8. Boge, Florian J. (2025). Models: Measuring or Cognitive Instruments? Journal for General Philosophy of Science. DOI 10.1007/s10838-025-09733-9. https://doi.org/10.1007/s10838-025-09733-9
  9. Malouchou Kanellopoulou, Chrysi Maria (2026). Is Farr's cholera law of elevation a counterexample to selective realism? Studies in History and Philosophy of Science 115:102105. DOI 10.1016/j.shpsa.2025.102105. https://doi.org/10.1016/j.shpsa.2025.102105