This book begins from a dissatisfaction familiar to many analytic philosophers: disputes about counterfactuals often reach an impasse because the parties report different intuitions about cleverly crafted cases. Huber does not join the game of defending his intuitions case by case. He rather relies on his unifying methodology of means–end philosophy by asking what the necessary means are to achieve the epistemological ends you may have. The key insight is that you can use these necessary means only if the world is a certain way. Means–end philosophy tells you what metaphysical theses you are committed to once you adopt certain epistemological ends. The result is an unusually original and systematic book: a work in which formal epistemology supplies hypothetical constraints on metaphysics.

You might wonder why you could not adopt epistemological ends while believing that the world is not set up to let you attain them. The answer turns on the kind of commitment at issue. Given your ends, you ought to satisfy certain norms. Yet you can satisfy these norms only if certain metaphysical theses hold. If you deny the theses, you hold that you ought to do what you cannot do. Nor is there a mind-independent modal reality that might fail to cooperate. On Huber’s modal idealism, the theses concern modality, and modality belongs not to reality but to the language, broadly construed, in which we represent reality. Such a language is not right or wrong, only more or less useful for various purposes, your epistemological ends among them.

The book is the second of two volumes on the project of means–end philosophy, containing chapters 7–10 and a conclusion. The ‘more’ in its title should be taken seriously: this is a continuation of the first, more introductory, volume (Huber 2021). The new philosophy is best understood after reading the first volume. This being said, the second volume is self-contained for readers familiar with ranking and probability theory.

Chapter 7 opens by arguing that intuitions about counterfactuals cannot decide between rival theories. The familiar disputes between Stalnaker (1968, 1981) and Lewis (1973a) over conditional excluded middle and between Gillies (2007) and Moss (2012) over reverse Sobel sequences become exhibits in a methodological case: agreement that an utterance that sounds odd does not determine whether its problem is semantic, pragmatic, or something else. Means–end philosophy, by contrast, has no need to unduly rely on intuitions. It only gets off the ground once an agent adopts an end or certain ends that fix what the problem is in the first place. Epistemological norms are understood as hypothetical imperatives. If you have certain ends, such as holding true and informative beliefs or avoiding error, you ought to take the necessary means to attain them. These ends can pull apart, as Huber notes. If you suspend judgement on every contingent proposition, you avoid error trivially. The royal rule, introduced below, is shown to be necessary and sufficient for typically shunning error; believing the truth requires further assumptions and principles. Huber concedes that intuitions are useful for arriving at such norms, but they do not justify them. The norms are justified by showing that the ends are indeed achieved by the means. The metaphysical theses are justified as necessary conditions: they must be true for the norms to be satisfiable and so for the ends to be attainable. Huber calls this metaphysics ‘transcendental’. Familiar transcendental arguments start from a fact about experience or thought that no one can deny and aim at categorical conclusions against the sceptic. Huber starts from ends you may or may not have, establishes the necessary conditions by proof, and draws conclusions as hypothetical as the norms.

The chapter then introduces Huber’s modal idealism and the royal rule, the norm at the centre of his epistemology. Modal idealism treats possible worlds as mind-dependent ideas, rather than the real entities of Lewis’s (1986) modal realism. For the modal idealist, modal propositions are claims about ideas. The modal idealist locates the modalities in the mind, as the expressivist does, but takes modal claims to be true or false, as the realist does. The modalities are not private, though. They belong to the language, in the broadest sense, in which we represent reality. Whether it is necessarily raining is therefore not settled by what anyone can imagine. It is settled by the language and by which factual world is actual. ‘Necessarily, it is raining’ is true at a world if, and only if, it rains at every world that this world’s atypicality function does not rank as infinitely atypical. The royal rule, modelled on Lewis’s (1980) principal principle, links descriptive atypicality to conditional grades of disbelief. Atypicality is purely descriptive and singular. It concerns how normal it is for this proposition to be true on this occasion and in this context, not how often things of its kind happen. Claims about it are empirical hypotheses and chapter 10 shows how they can be tested. The royal rule is very roughly this: an ideal agent, Ida, should disbelieve a factual proposition A to degree n given the information that A is atypical to degree n, as long as she has no further information that overrides it. In his first volume, Huber argues that beliefs ought to obey the ranking calculus. Assuming this together with his semantic clauses, the royal rule determines a logic for default conditionals, as well as one for counterfactuals.

Let’s illustrate Huber’s means–end methodology. Suppose ranking-theoretic agent Ida aims to typically shun error and is certain of nothing but the tautology. Given the information that rain tomorrow is atypical to degree two, following the royal rule requires her to conditionally disbelieve rain to degree two, that is, to conditionally believe that it will not rain tomorrow. The qualification matters because the royal rule lapses when Ida has inadmissible information, such as direct evidence about whether it will rain. Following the recommendation of the royal rule is thus a means to her attaining the end of typically shunning error. The rule itself constrains the structure of atypicality by rational grades of disbelief in a way that is similar to the principal principle’s constraint of the structure of chances by rational credences. Given Huber’s semantic clauses, the atypicality structure determines the logical principles of default conditionals and counterfactuals. The argumentative direction is revolutionary. Rather than consult intuitions and cases to select the counterfactual logic, Huber begins with the epistemic end of typically shunning error, identifies a necessary means for attaining it, which results in the royal rule, and finally derives from this means–end norm the logical postulates of default conditionals and counterfactuals as metaphysical theses.

The means–end methodology shows that a metaphysical structure is required by a particular package of epistemological ends and semantics. Such a conditional stance is legitimate within an instrumentalist project. And Huber states his assumptions candidly: counterfactuals express truth-valued propositions at possible worlds, descriptive typicality supplies their central semantic ingredient, and so on. Readers who accept these premises receive a striking derivation. However, one may wonder what the dialectical reach of the resulting metaphysics is. How can it engage, or even persuade, someone who disagrees with the ends or the semantics? What if I want to shun error—not merely typically so?

Chapter 8 applies the means–end methodology to metaphysics. Its first section justifies the royal rule by showing that following it is a necessary and sufficient means for typically shunning error. To get a glimpse, let’s say Ida has this end, believes that, typically, Tweety can fly if it is a bird, and learns that Tweety is a bird. Her belief concerns Tweety, not birds in general. By the royal rule, Ida should disbelieve that Tweety cannot fly to the degree that this proposition is atypical, and so should believe that Tweety can fly. Since she follows the rule, a situation in which Tweety can fly but Ida believes otherwise is atypical and the more atypical it is, the firmer her belief. If Ida correctly believes what is typically the case, she typically correctly believes what is the case. Of course, had Ida learned that Tweety is an ostrich, she could not have used the rule that birds typically fly because the information about Tweety contradicts the claim that it can fly. It is typical for the ostrich Tweety that it cannot fly. In this sense, following the royal rule is for Ida nothing but typically shunning error.

The next section applies the modal idealist’s notion of descriptive typicality to problems in the philosophy of actual causation. While Huber does not adopt a theory of actual causation, the result allows for a theory that unifies deterministic and probabilistic causation. However, this feat comes at a cost: the unified theory makes counterfactual dependence sufficient for actual causation, as it is on Lewis’s (1973b) account.  If Putin had watered my plant, it would not have died. True—but Putin’s omission is arguably not a cause of my plant’s death.¹

In the last section of the chapter, Huber applies Arrow’s (1951) impossibility theorem for social choice to Lewis’s (1979) determination of overall similarity between possible worlds. The determination of overall similarity by individual aspects of similarity among possible worlds—such as violation of laws of nature and match in particular matters of fact—has the same structure as the aggregation of individual preferences into an overall group preference. Hence, Arrow’s theorem entails that the system of weights Lewis wants to use for the determination cannot yield a relation of overall similarity over possible worlds. The aggregation problem can be illustrated thus: one merely possible world might preserve more of the actual world’s particular facts but require a small violation of a law; another might obey every law but depart more extensively from particular facts. Lewis needs an overall verdict assembled from these distinct comparisons. Arrow’s theorem shows why no such aggregation rule exists. Numerical ranks provide one escape from the impossibility theorem because they contain more information than ordinal comparisons (Kroedel and Huber 2013). Huber ultimately recommends a deeper revision: possible worlds should be ordered by where violations occur in a causal hierarchy, not by a vague combination of law violations and match of particular facts. For Huber rejects not only the best-system analysis but the very notion of a law of nature. This discussion is compact, accessible, convincing, and makes a big step forward in our understanding of causal counterfactuals.

Chapter 9 develops formal frameworks to characterize interventions, as well as causal and backtracking counterfactuals. Huber embeds Halpern’s (2016) extended structural-equation models in ‘typicality models’. He shows that the distinction between exogenous and endogenous variables embodies causal information, and how structural equations can be reproduced through default conditionals and world-relative rankings. A further generalization leads to ‘causality models’, in which causal laws and even the exogenous–endogenous division can vary across worlds, allowing for truth conditions of left-nested interventionist counterfactuals. Several representation theorems establish the relations among these frameworks. If you are looking for formal philosophical insights, look no further.

The payoff is a precise account of the difference between backtracking and causal counterfactuals. Backtracking counterfactuals hold fixed the whole causal structure and nothing but. Causal counterfactuals hold fixed the causal structure and are ‘focused on actuality’, which means roughly that the exogenous variables take on their actual values. Notably, this treatment renders soft and hard interventions definable, while others need to assume them as primitives.² It also revises Lewis’s weights: early violations in the causal hierarchy matter infinitely more than downstream ones for both types of counterfactuals, exogenous matches of fact matter only for causal counterfactuals, and endogenous ones for none. This answers Fine’s (1975) future similarity objection to Lewis’s semantics: ‘if Nixon had pressed the button, there would have been a nuclear holocaust’ comes out true. Readers interested in the work of Spirtes et al. (2000), Pearl (2009), Halpern (2016), Hitchcock (2001, 2007), Woodward (2003), or Spohn (2012) will find much to engage with here. Readers seeking a discursive metaphysics of causation may find the density of definitions, constraints, and proofs a bit hard to digest.

To get a feel of the characterizations, consider Huber’s recurring example. Ida drinks wine, sleeps in, and goes running. Her drinking wine causally determines both her sleeping in and her running. Suppose that she did not sleep in. On a backtracking reading, we retain the whole causal structure and revises its upstream conditions: if Ida had not slept in, she must not have drunk wine and consequently would not have gone running. On a causal reading, not sleeping in is treated as the result of an intervention. The upstream fact that Ida drank wine is held fixed because the intervention breaks the connection between drinking and sleeping in; and so she still goes running. Huber generalizes this contrast. Backtracking counterfactuals hold fixed the causal structure and nothing else. Causal counterfactuals additionally respect actuality, roughly, by holding fixed the actual values of exogenous variables and thereby what is not causally downstream of the antecedent.³ This characterization explains how the same antecedent can support incompatible counterfactual consequents on the different readings.

Chapter 10 asks how counterfactuals can be empirically tested even though merely possible worlds cannot be inspected or sampled. Huber first argues that intervention is a form of conditioning on richer evidence: making the antecedent true carries more information than observing that it is true. Merely observing that Ida did not sleep in supports an inference to a change in its cause—she did not have wine. Intervening to prevent her from sleeping in supplies the information that the antecedent was made true by the intervention—she still had wine the night before. The intervention carries the additional information that everything not causally downstream of her sleeping in remains how it actually was. Huber’s semantic distinction of backtracking and causal counterfactuals in terms of what information is conditioned on entails that causal decision theory can be seen as a form of evidential decision theory respecting Carnap’s (1947) principle of total evidence. The distinction also allows the ranking theorist to empirically test counterfactuals in analogy to probabilistic testing. A generalization of what statisticians call the ‘modes’ of a sample plays the role of frequencies; the royal rule plays the role of the principal principle; an ‘obvious observation’ plays the role of the strong law of large numbers. Under independence and consistency assumptions, observed modes can support hypotheses about typicality and so about counterfactual claims.

We have seen many virtues of Huber’s means–end philosophy. However, its methodology risks avoiding rather than settling disputes. It seems that the selection of epistemological ends and semantic assumptions requires more careful defence. But Huber does not tell us what ultimate ends an agent should have, only what follows from certain ends an agent may, or may not, adopt. This reservation merely anticipates where the ambitious unification project is most likely to provoke productive disagreement. By no means does it diminish the book’s value as a research contribution. We should not forget that Huber offers a revolutionary philosophical methodology, complete with formal machinery, applications, and a prospectus for ranking-theoretic decision by anticipated regret (see the conclusion). The reader is, of course, free to reject modal idealism, resist the royal rule, or doubt the specific end of typically shunning error. But she cannot unlearn that Huber’s system, and any comparable epistemology equipping rational agents with ends, entails metaphysical commitments.

Who, then, should read it? Specialists in formal epistemology, counterfactual logic, causal modelling, and the metaphysics of causation should. Chapter 9 does not require any knowledge of ranking theory and is especially valuable for philosophers interested in the relation between structural equations and possible-worlds semantics. Those concerned with confirmation and causal inference should read chapter 10. It is no introduction, but for readers interested in whether ranking theory can carry an extraordinary explanatory load, it is original, rigorous, and consistently stimulating. Read together with the first volume, it presents a distinctive answer to an intriguing question: how far can philosophy proceed by deriving what we must assume from what we aim to achieve?

Mario Günther
Panthéon Sorbonne
Mario.Gunther@univ-paris1.fr

Notes

¹ See (Andreas and Günther 2025a, 2025b, 2026).

² For example, (Spirtes et al. 2000; Woodward 2003; Pearl 2009).

³ For a similar characterization of causal conditionals, see (Günther 2022).

References

Andreas, H. and Günther, M. (2025a). ‘Factual Difference-Making’, Australasian Philosophical Review, 9, pp. 118–48.

Andreas, H. and Günther, M. (2025b). ‘Reply to Commentaries on “Factual Difference-Making”’, Australasian Philosophical Review, 9, pp. 213–37.

Andreas, H. and Günther, M. (2026). From Reasons to Causes: A Theory of Causation, Cambridge University Press.

Arrow, K. J. (1951). Social Choice and Individual Values, Wiley.

Carnap, R. (1947). ‘On the Application of Inductive Logic’, Philosophy and Phenomenological Research, 8, pp. 133–48.

Fine, K. (1975). ‘Critical Notice (Review of Counterfactuals by David K. Lewis)’, Mind, 84, pp.  451–58.

Gillies, A. S. (2007). ‘Counterfactual Scorekeeping’, Linguistics and Philosophy, 30, pp. 329–60.

Günther, M. (2022). ‘Causal and Evidential Conditionals’, Minds and Machines, 32, pp. 613–26.

Halpern, J. Y. (2016). Actual Causality, MIT Press.

Hitchcock, C. (2001). ‘The Intransitivity of Causation Revealed in Equations and Graphs’, Journal of Philosophy, 98, pp. 273–99.

Hitchcock, C. (2007). ‘Prevention, Preemption, and the Principle of Sufficient Reason’, Philosophical Review, 116, pp. 495–532.

Huber, F. (2021). Belief and Counterfactuals: A Study in Means–End Philosophy, Oxford University Press.

Kroedel, T. and Huber, F. (2013). ‘Counterfactual Dependence and Arrow’, Noûs, 47, pp. 453–66.

Lewis, D. K. (1973a). Counterfactuals, Harvard University Press.

Lewis, D. K. (1973b). ‘Causation’, Journal of Philosophy, 70, pp. 556–67.

Lewis, D. K. (1979). ‘Counterfactual Dependence and Time’s Arrow’, Noûs, 13, pp. 455–76.

Lewis, D. K. (1980). ‘A Subjectivist’s Guide to Objective Chance’, in R. C. Jeffrey (ed.), Studies in Inductive Logic and Probability, Vol. 2, University of Berkeley Press, pp. 263–93.

Lewis, D. K. (1986). On the Plurality of Worlds, Blackwell.

Moss, S. (2012). ‘On the Pragmatics of Counterfactuals’, Noûs, 46, pp. 561–86.

Pearl, J. (2009). Causality: Models, Reasoning, and Inference, Cambridge University Press.

Spirtes, P., Glymour, C. and Scheines, R. (2000). Causation, Prediction, and Search, MIT Press.

Spohn, W. (2012). The Laws of Belief: Ranking Theory and Its Philosophical Applications, Oxford University Press.

Stalnaker, R. C. (1968). ‘A Theory of Conditionals’, in N. Rescher (ed.), Studies in Logical Theory, Blackwell, pp. 98–112.

Stalnaker, R. C. (1981). ‘A Defense of Conditional Excluded Middle’, in W. Harper, R. Stalnaker, and G. Pearce (eds), Ifs: Conditionals, Belief, Decision, Chance, and Time, D. Reidel, pp. 87–104.

Woodward, J. (2003). Making Things Happen: A Theory of Causal Explanation, Oxford University Press.