Bayesian Reasoning

The free energy principle holds that any system persisting across a statistical boundary can be described as though it minimizes a single quantity, the variational free energy, and it has been offered as a unifying account of mind and life. Its status is contested: critics have questioned whether the Markov blanket construction is well defined and whether the principle explains anything or merely redescribes it. This paper brings a different kind of test to that debate. Because the variational free energy (...) is built from probabilities it is dimensionless, and dimensional analysis supplies a unit-independent, theory-neutral criterion for physical content: a quantity that genuinely couples to the world must appear as a function of the physical dimensionless groups of the system, whereas a quantity whose value can be changed without moving any such group is not tracking the physics. Applied in a concrete, physically bounded system, bacterial concentration sensing, whose Markov blanket is an actual membrane, the criterion shows that the variational free energy partitions into three parts: a leading term equal to a known physical quantity, the Berg-Purcell measurement number; a dominant correction that is independent of every physical group and can be moved at will by the agent's priors; and a subleading term that does couple to the physics but vanishes as sensing improves. In this system the principle is a bookkeeping device laid over a coupling law that dimensional analysis exhibits directly. The test generalizes: any postulated coupling quantity can be asked whether its dimensional content collapses onto a physical group or floats free of them all. (shrink)

Beliefs can be more or less supported by evidence. When the evidence speaks in favour of some proposition p, you can rationally be confident that p. On the Bayesian interpretation of doxastic states, this amounts to a high credence in p. But this cannot be the full account of confidence. Gaining evidence that supports the credence you already have improves your epistemic situation. Here, your credence level should remain unchanged, but the credence should become more resistant to change. Doxastic resilience (...) measures your disposition to change credence in the face of incoming evidence. -/- The central claim of this thesis is that by thinking of confidence in terms of both the level and resilience of credences, we can make progress on several philosophical problems. After briefly introducing doxastic resilience in Chapter 1, I put the concept to use in four philosophical debates. -/- Chapter 2 offers an evidential account of rational resistance to counterevidence which contrasts with views according to which knowledge plays a special role in safeguarding against counterevidence. Chapter 3 argues that loss of resilience can influence rational decision-making, which sheds light on the normative significance of certain kinds of higher-order defeat. This contrasts with imprecise probability views on which higher-order defeat increases the imprecision of credence. Chapter 4 argues that higher-order uncertainty calibrates first-order credences and constrains expected resilience. This explains both higher-order confirmation and defeat. Chapter 5 shows that resilience bears on the expected value of inquiry. When unresilient, one should expect new evidence to be valuable. This explains cases where seemingly identical decision-relevant credences and utilities should nevertheless make different decisions when inquiry is feasible. (shrink)

Philosophical arguments for the rationality of Bayesian conditioning often use assumptions and concepts from epistemic logic without fully explicating them. In this work, we present a general dynamic epistemic logic framework to analyze logical properties of (Bayesian) learning scenarios. We argue that Perfect-Recall is an important epistemic principle at the heart of the project of justifying Bayesian conditioning. In such epistemic logic formulation, Perfect-Recall is not really about memory, but about the agent’s general ability to know how they came to (...) know what they know. Following the existing literature, we use Monty Hall style cases to demonstrate the connection between PerfectRecall and the commitments of Bayesian conditioning. (shrink)

Building on Nozick’s (2001) work, I provided a novel invariantist account of scientific objectivity: the Counterfactual Independence Account (Reutlinger 2024). Despite its virtues, this account needs to be articulated in a more nuanced manner – regarding its scope, content, and consequences. My goal in this paper is to develop such a more articulated and improved version of the Counterfactual Independence Account in two steps. First, I will discuss two instructive dissonances between Nozick’s version of invariantism and the Counterfactual Independence Account (...) – dissonances concerning what concept of objectivity is addressed and how the key notion of invariance is characterized. Second, I will develop an argument for the claim that (epistemic) objectivity plays a productive role in scientific inference or, more precisely, that objective evidence is, in a sense to be specified, more evidence. (shrink)

Philosophers often speak of normativity as if it named a single phenomenon requiring a unified metaphysical explanation. This paper argues that the term is systematically overburdened. In discussions of moral ontology, normativity usually refers to categorical practical authority: obligations, requirements, or reasons that purport to bind agents independently of their contingent desires, projects, values, or commitments. In discussions of rationality, logic, epistemology, and decision theory, however, normativity usually refers to structural coherence: requirements internal to an agent's beliefs, ends, standards, methods, (...) or avowed commitments. These are not merely two applications of one deep concept. They have different logical forms. Moral normativity purports to evaluate the content of an agent's commitments from a standpoint independent of those commitments. Rational normativity evaluates the consistency of commitments from within a standpoint already adopted. A basketball player who claims to want victory while deliberately scoring for the opposing team illustrates the distinction. The player has acted incoherently relative to his stated aim, but he has not thereby acted immorally. His error is not moral wickedness but practical inconsistency. The paper also argues that the denotation of normativity can be clarified by attending to its etymological neighbors and antonyms: normal, abnormal, and abandoning norms suggest deviation from a standard, pattern, rule, or expectation, not automatic moral wrongness. The paper argues that much philosophical confusion results from treating normativity as a free-standing noun rather than a parameterized relation: normative for whom, relative to which standard, in which domain, with what force, and conditional on what? Once these parameters are made explicit, the existence of rational, logical, or instrumental normativity gives no independent support to moral realism. The moral non-realist can accept all the normativity required for logic, evidence, and practical consistency while rejecting stance-independent moral facts. (shrink)

Calling the Big Bang a "miracle" can be rhetorically powerful because the term carries both wonder and theological agency. This paper argues that the argument often depends on equivocation. If miracle means an astonishing or currently unexplained event, the Big Bang may be miracle-like only in a loose aesthetic sense. If miracle means an act of God that violates or transcends natural order, then the theological conclusion has been built into the premise. The paper distinguishes wonder, anomaly, boundary condition, divine (...) act, and evidential sign. It then shows how apologetic rhetoric slides among these meanings to convert cosmological mystery into theistic confirmation. The result is not a denial that cosmology raises profound metaphysical questions. It is a refusal to let semantic inflation do the work of evidence. (shrink)

Kierkegaard's king-and-maiden motif is often used to illuminate divine hiddenness: a king who wants genuine love may approach the beloved in disguise rather than overwhelm her by royal glory. This paper argues that the analogy is frequently asked to do more than it can. It may explain why coercive spectacle would be relationally defective, but it does not explain why recognition should be so difficult, uneven, and historically mediated. Love requires freedom, but it also requires access to the lover. Incognito (...) can protect authenticity only if it remains a route to encounter rather than a barrier to identification. Applied to divine hiddenness, the analogy therefore supports a narrower conclusion than many defenders need. It distinguishes non-coercion from non-recognition and exposes a gap between preserving free response and leaving sincere seekers without adequate relational access. (shrink)

The claim that "we have never seen something come from nothing" is often used to support a causal premise in cosmological apologetics. This paper argues that the move is an inductive cherry. It selects one ordinary-world induction while ignoring other ordinary-world inductions that cut against the desired conclusion: minds are embodied, agents act through mechanisms, causes operate within temporal orders, complex intentions arise from prior processes, and explanations that multiply unobserved powers lose predictive discipline. The paper does not argue that (...) induction refutes every cosmological argument. It argues that selective induction cannot bear the burden placed on it. If ordinary experience is allowed into cosmic-origin reasoning, it must enter as a portfolio rather than a single convenient slogan. A more comprehensive inductive audit weakens the inference from "not from nothing" to a timeless, immaterial, personal creator. (shrink)

We probabilify the scenario approach to legal proof. The scenario approach searches for the scenario that strikes the best balance in explaining the available evidence, in fitting to general background beliefs, and in its degree of internal coherence. Our account provides a unified measure of the three dimensions in terms of probabilities, and so is proof that the scenario approach can be probabilified. Indeed, our account can be summarized by a version of Bayes Theorem: the most likely scenario in light (...) of the evidence is the best. We thereby provide a Bayesian justification for the scenario approach. (shrink)

Romans 1:20 is often read as a sweeping epistemic claim: God's invisible qualities, eternal power, and divine nature are clearly perceived from creation, leaving human beings without excuse. This paper asks what public evidence should exist if that claim is true in the robust apologetic sense. If nature clearly discloses the relevant God to ordinary human observers, we should expect a broad archive of independent testimonies in which people across cultures report arriving at the same divine target through creation itself. (...) The observed record looks different. Human beings do infer agency, purpose, sacred presence, spirits, gods, ancestors, cosmic order, and moralized surveillance from nature. But these inferences diversify dramatically. Creation has been read as pointing to Jupiter, Ahura Mazda, Allah, Brahman, ancestral spirits, deism, pantheism, impersonal order, and countless local powers. The pattern is not clear perception of one determinate divine nature; it is underdetermined religious projection filtered through culture, cognition, and inherited categories. I develop a testimony-gap argument. The problem is not that no one has ever seen divinity in nature. The problem is that nature does not publicly discriminate among rival religious interpretations with the clarity Romans 1:20 is often made to claim. If the verse is reduced to the claim that humans can form vague theistic or spiritual impressions, it loses much of its apologetic force. If it is kept strong enough to ground culpability, the missing cross-cultural archive becomes evidentially serious. (shrink)

This paper develops a methodological critique of hiddenness defenses. It is distinct from arguments that clearer evidence would not coerce free allegiance and from arguments that divine attributes generate expectations of accessibility. Those companion arguments ask what clarity would do and what love should predict. The present paper asks how popular defenses of divine hiddenness move after pressure is applied. The central problem is elasticity: the explanation offered for divine absence changes shape according to the local objection while preserving the (...) same conclusion that God remains unsurprisingly hidden. The critique is not that every theistic reply to hiddenness is foolish. Serious philosophers have offered sophisticated accounts involving freedom, morally significant seeking, divine pedagogy, spiritual transformation, and noncoercive self-disclosure. The narrower claim is that many hiddenness defenses lack constraint conditions. They do not specify in advance which pattern of divine silence would count against them, which distribution of nonbelief would be surprising, which sincere seeker would be evidentially relevant, or what degree of ambiguity would be excessive. I frame the problem in terms of model mobility, ad hoc rescue, and evidential risk. A defense that can explain divine clarity, divine ambiguity, lifelong nonbelief, demographic clustering, doctrinal confusion, unanswered seeking, and every intermediate case has become too flexible to confirm anything. It may protect belief psychologically, but it no longer functions as a disciplined explanation. The conclusion is modest: if hiddenness defenses are to be more than theological shock absorbers, they must name the patterns they would not expect. (shrink)

This paper strengthens a triviality result by Santorio (2022, Analysis) for the probability of counterfactuals. The best response may be to drop the idea that the probability of counterfactuals behaves classically.

This paper develops a disciplined use of anthropic reasoning against its ideological misuse. The anthropic principle properly reminds us that observations are conditioned by the existence of observers. We observe a life-permitting environment because only such environments can contain observers like us. That conditional fact can correct naive surprise. It does not by itself show that the universe exists for us, that human beings are cosmically central, or that fine-tuning is best explained by divine intention. The contribution is not another (...) general critique of design arguments. The focus is observer-selection humility: the epistemic posture required when survival and observation are preconditions for noticing anything at all. Survivors of low-probability filters are tempted to interpret their presence as purpose. Anthropic reasoning should resist that temptation, not sanctify it. Drawing on Carter, Barrow and Tipler, Bostrom, Leslie, Sober, Collins, and White, I distinguish selection effects from teleological promotion. Fine-tuning arguments may still raise serious questions, but they must compare hypotheses carefully. The datum "we observe a life-permitting universe" is not neutral evidence available before conditioning on observers. A formal section shows why \(P(observation observer exists)\) cannot be straightforwardly converted into \(P(design observation)\). The conclusion is that anthropic reasoning should produce humility about our reference class, not confidence that humanity is the point of the cosmos. (shrink)

This paper critiques a pragmatic apologetic strategy associated with William Lane Craig: the suggestion that even a tiny probability of Christianity may be enough to make belief worthwhile because the promised benefits are so great. The problem is not that practical stakes are irrelevant to inquiry. High stakes can justify investigation, caution, or provisional action. The problem is the conversion of pragmatic payoff into epistemic permission to believe. A low probability attached to a vast reward may affect decision theory; it (...) does not raise the truth probability of the proposition. The argument distinguishes four attitudes often blurred in popular apologetics: belief, investigation, hope, and wager-like action. A one-in-a-million possibility might justify examining a claim if the cost is low and the payoff is high. It might justify hope or precaution. It does not justify believing the claim true. To believe on such a basis is to lower the evidential bar precisely where the claim is most extraordinary. Drawing on Pascal, James, Clifford, Hájek, Craig's broader apologetic work, and Bayesian decision theory, I argue that the millionth-chance standard is unstable. If applied symmetrically, it licenses rival religious commitments with enormous promised payoffs. If applied only to Christianity, it is special pleading. If defended as pragmatic rather than epistemic, it no longer supports belief. A formal section separates expected utility from credence and shows why practical benefits cannot substitute for evidence. (shrink)

This paper analyzes the claim that faith is unavoidable. The claim is usually meant to protect religious faith from criticism by placing it in the same category as ordinary trust in bridges, aircraft, memory, testimony, induction, and expert practice. I argue that this maneuver depends on faith inflation: the expansion of faith until it includes any reliance under uncertainty. Once inflated, faith becomes universal but trivial. When narrowed back to belief or commitment that outruns evidential support, faith is no longer (...) unavoidable. The paper is a sister treatment to The Faith-Reversal Trap, which criticized the slogan that atheism requires more faith than Christianity. The present paper targets a different move: not reversal, but inevitability. Its central distinction is between rational confidence and evidential excess. Rational confidence is defeasible, proportioned to track record, and sensitive to counterevidence. Faith, in the relevant epistemic sense, begins where confidence exceeds what the evidence warrants. Drawing on Clifford, James, Buchak, Bayesian epistemology, and Reformed epistemology, I argue that religious faith is not forced by the ordinary structure of human life. We must rely on uncertain evidence; we need not inflate credence beyond it. A compact formalization defines faith as \(Cr(p) > Ev(p)\), where personal confidence in \(p\) exceeds evidential warrant for \(p\). On that definition, faith is not a universal condition. It is an optional epistemic posture, sometimes existentially understandable, but not made rational by calling it unavoidable. (shrink)

Believers often claim that evidence for God is all around us: in cosmic order, beauty, moral experience, consciousness, fine-tuning, providence, religious experience, and the intelligibility of nature. This paper argues that such evidence, even when granted for argument's sake, usually suffers from a specificity failure. The data invoked are not distinctively Christian. They are recruited by many mutually incompatible religions, by generic theism, by deism, by pantheism, and by spiritual but non-doctrinal worldviews. Evidence that is equally available to rival claimants (...) cannot by itself confirm one claimant in particular. The critique does not deny that natural theology can raise interesting questions. Hume's design critique, Swinburne's cumulative case, and contemporary debates over explanation show that order, consciousness, and religious experience are philosophically serious topics. The narrower issue is evidential routing. What exactly does a datum support? A feature of the world might support thin theism more than naturalism, while still failing to support Christianity over Islam, Hinduism, Sikhism, deism, or other theistic hypotheses. I develop the generic evidence problem through confirmation theory, underdetermination, religious diversity, and rival uptake. A formal section shows that evidence E confirms hypothesis H_1 over H_2 only when E is more expected under H_1 than under H_2. If the same evidence is expected under many mutually exclusive religious hypotheses, then it has little power to select among them. The paper concludes that apologetics must stop treating generic compatibility as specific confirmation. (shrink)

The slogan that it takes more faith to be an atheist has become a durable apologetic reversal. Popularized in contemporary evangelical culture by works such as Geisler and Turek's I Don't Have Enough Faith to Be an Atheist, the slogan attempts to turn a perceived weakness of religion into a weakness of unbelief. This paper argues that the reversal succeeds rhetorically by exploiting semantic instability in the word faith. If faith is a virtue, accusing atheists of having more of it (...) is not a criticism. If faith is a defect, then the apologist has undermined the religious posture he elsewhere praises. The argument is not that all Christian uses of faith are irrational or that all atheists reason well. The narrower claim is that the slogan depends on switching between faith as trust, faith as confidence under uncertainty, faith as commitment, and faith as belief beyond sufficient evidence. Once those meanings are separated, the slogan loses its force. Atheism may involve uncertainty, assumptions, trust in testimony, or confidence in scientific institutions. None of that makes it faith in the sense relevant to Christian apologetic accusation. Drawing on evidentialism, philosophy of faith, and work on motivated reasoning, I show that the slogan performs three tasks: it reverses the burden of proof, normalizes religious faith by universalizing it, and then stigmatizes atheism for allegedly possessing the very feature religion sanctifies. A compact formalization clarifies the contradiction. The paper concludes that serious inquiry should replace faith-inflation rhetoric with explicit talk of credence, evidence, trust, and practical commitment. (shrink)

This paper examines a common apologetic use of the problem of induction. Hume's challenge shows that inductive expectation cannot be justified by deduction without circularity. Christian apologists sometimes convert that shared limitation into an argument that non-theistic rationality is unstable while theism secures induction through divine order, providence, or covenantal faithfulness. I argue that this is an ownership fallacy. Theism does not solve induction merely by naming a divine guarantor, and non-theistic inquiry does not collapse because it lacks a deductive (...) proof of uniformity. The paper is deliberately distinct from two nearby treatments. Induction: The Best Game in Town defended induction by performance, corrigibility, and pragmatic vindication. Induction Before Deduction argued that human use of deduction depends on inductive learning and validation. The present paper addresses a different target: the apologetic attempt to claim that Hume's problem gives theism a proprietary advantage over secular reasoning. I develop the critique through Hume, Reichenbach, Goodman, Quine, Norton, and contemporary confirmation theory. The central point is that a theological premise such as "God orders the world" does not bridge the epistemic gap unless we already have reliable access to God's intentions, know that divine freedom will preserve the relevant regularities, and can distinguish genuine providential order from our inductively learned expectations. Those bridge claims are themselves inductive, interpretive, and defeasible. The conclusion is that the problem of induction is a discipline on all finite inquiry, not a trophy that any worldview may seize by stipulation. (shrink)

Christian apologetics often begins with a psychologically powerful observation: human beings experience the world as broken. We feel mismatch, loss, injustice, moral frustration, unmet longing, and the ache of possible improvement. This paper argues that the apologetic move from that unease to Christian doctrine is not an argument so much as an affective conversion device. It takes a broadly human phenomenon-the capacity to compare what is with what might be-and treats it as though it were evidence for a specific theological (...) story of fall, sin, redemption, and divine restoration. The critique is not that the world contains no suffering, disorder, or moral failure. It plainly does. Nor is it that religious narratives cannot provide meaning to those who inhabit them. The narrower claim is that existential discontent is evidentially underdetermined. Alienation, grief, moral aspiration, and dissatisfaction are compatible with many accounts: evolutionary psychology, social comparison, moral imagination, political critique, Buddhist dukkha, tragic humanism, secular progressivism, Christian fallenness, and other religious cosmologies. The apologetic argument succeeds rhetorically by moving too quickly from a shared mood to a sectarian diagnosis. I develop the argument through existential philosophy, sociology of religion, cognitive bias research, and confirmation theory. The central distinction is between diagnostic resonance and evidential discrimination. A story may resonate because it gives pain a name, but resonance does not show that the story is true. The paper concludes that broken-world apologetics should be treated as pastoral rhetoric unless it can specify why human discontent confirms Christian doctrine more strongly than rival interpretations of the same data. (shrink)

Resurrection apologetics often claims that the bodily resurrection of Jesus is the best or most probable explanation of the empty tomb, post-mortem appearance reports, and the rise of early Christian conviction. This paper argues that such language is frequently stronger than the probabilistic work being done. A comparative explanation is not entitled to the phrase "most probable" unless the relevant prior probabilities, likelihoods, rival hypotheses, and hypothesis space have been stated with enough clarity to make the comparison meaningful. Otherwise the (...) apologetic argument substitutes narrative preference for probabilistic superiority. The central problem is not that every historical judgment must use precise decimal values. Historians often reason with qualitative comparisons, rough likelihoods, and defeasible judgments. The problem is that resurrection apologetics commonly makes a maximal comparative claim while refusing the burdens that make such a claim responsible. A miracle hypothesis cannot be allowed to remain probability-free while its rivals are pressed for defects. Nor can the apologetic case count the explanatory power of resurrection without also confronting the prior improbability of a dead body returning to embodied life. I develop the critique through Bayesian confirmation theory, Humean miracle skepticism, and contemporary work on resurrection arguments. I also offer a compact formalization showing why "best explanation" rhetoric fails when the miracle prior is hidden. The result is a narrower but sharper objection than a generic rejection of miracles: if the resurrection is to be defended as the most probable explanation, its defenders must state the probability terms they are using rather than treating probability as a rhetorical honorific. (shrink)

Conciliationism holds that it is rational to modify one's beliefs in the face of disagreement. But extant conciliatory norms yield incorrect results in cases involving excessive confidence--cases in which one's interlocutors are sure, or almost sure, that their opinions are correct. After explaining the problem of excessive confidence, I show that a Bayesian approach to Conciliationism handles the problem elegantly and effectively. Further, it has desirable--indeed essential--features, including the ability to (i) contend with multiple interlocutors, (ii) deal with gradations in (...) competence, (iii) incorporate ubiquitous interdependence, and (iv) account for critical contextual features of cases of disagreement. (shrink)

Sean Carroll's criticism of theism in the 2014 God and Cosmology debate was not merely that theism lacks a mechanism or that naturalism currently has better physics. His sharper point was that theism has too much accommodation capacity. When a hypothesis can reinterpret almost any possible observation as compatible with divine intention, it loses the ability to discriminate among possible worlds. This paper develops that point as a general account of explanatory elasticity. A theory becomes elastically weak when its auxiliary (...) resources allow it to stretch around data without making risky, specific, antecedent expectations. The paper situates Carroll's point within philosophy of science, Bayesian confirmation, and model selection. Popper emphasized risk and falsifiability. Quine showed that hypotheses face evidence within a web of auxiliaries. Sober and Forster argue that simplicity and predictive accuracy are connected through constraints on accommodation. I argue that theism often treats auxiliary elasticity as a virtue: God might want life, beauty, discoverability, moral drama, hiddenness, lawlike order, miracles, simplicity, complexity, suffering, or rescue. But the more possible worlds theism can bless after the fact, the less evidential work any particular world performs. I formalize the problem with a flexibility penalty. A hypothesis (H) confirms poorly when its auxiliary set (A_H) can absorb a wide range of mutually different observations while offering few independent constraints. Theistic explanations frequently raise (P(E mid H,A_H)) only by expanding (A_H) after (E) is known. That strategy may preserve logical consistency, but it reduces predictive discipline. The paper concludes that Carroll's critique is best understood not as scientistic impatience with theology, but as a demand that cosmological explanations pay for their elasticity. (shrink)

Philosophical discussion of the two-envelope paradox has suffered from a lack of formal precision. I discuss various versions of the paradoxical argument using formal probability theory, which helps to make diagnoses that are simpler, more insightful, and provably correct. Paradoxical arguments are revealed to be fallacious for one of three reasons: (1) the argument makes a formal mistake such as an equivocation fallacy; (2) the argument disregards relevant uncertainty about or variability in a unit of measurement; (3) the argument uses (...) an invalid decision rule. I improve upon various existing diagnoses and discuss what kind of philosophical and decision-theoretic import the paradox has. (shrink)

This paper aims at an integrated account of truth, probability and acceptance conditions of counterfactuals. The basic idea is that truth conditions are an abstract semantic device: they are not directly tested against semantic judgments, but determine the probability of a sentence, and via probabilistic acceptance conditions, they generate predictions of semantic judgments. The proposed account has two notable features: first, the probability of counterfactuals can be understood in a purely subjective sense, without reference to objective chance, and second, Lewis-style (...) truth conditions for counterfactuals that quantify over multiple scenarios can be interpreted as describing their acceptance conditions. (shrink)

Bayesian conditionalization is rigid: learning E fixes p(E) at 1 while preserving probabilities conditional on E. Non-rigid update is preferable when, in the course of learning that E is true, we change our views about how—by way of which truthmakers ϵ. A Jeffrey-style generalization of Bayes—active conditioning—is developed which gives learning events a handle on p(ϵ|E) and p(E) both. E brings a truthmaker-incorporating “probasition” to the table, rather than simply an intension. Confirmation relations go hyperintensional as a result. Eis true (...) in the same worlds may not license the same updates, if their truth flows from different sources. (shrink)

We often treat observable results as if they directly determine the properties they are taken to indicate. A model performs well, and we attribute intelligence; a signal is detected, and we infer a property. This paper argues that such inferences are not merely epistemically risky but structurally underdetermined. -/- Observable results do not uniquely determine their interpretation. The same result may arise from multiple incompatible underlying structures, and no structure-preserving principle internal to the system selects between them. Interpretation is therefore (...) not only uncertain but formally underdetermined. -/- The source of this indeterminacy is identified as a systematic conflation of two distinct operations: representation (concept → measurement) and attribution (measurement → concept). To formalize this distinction, the paper introduces a minimal framework of epistemic registers—scientific, philosophical, and metaphysical—connected by type-constrained transformations. Representation is shown to be structure-preserving, whereas attribution is not. -/- This asymmetry propagates into probabilistic inference, decision theory, and proxy-based optimization, where it manifests as instability, divergence, and Goodhart-like failures. Artificial intelligence provides a limiting case in which this structure becomes particularly visible. -/- The central claim is that, in the absence of explicit constraints on transitions between epistemic registers, interpretation lacks structural determinacy within the system. (shrink)

The Monty Hall problem is a problem for causal decision theorists. Or so I argue here. My case for this claim is based on a novel "exotic" variant of the original Monty Hall problem, wherein the game-show contestant has foreknowledge of what door the game-show host will open. I argue that the causal decision theorist's treatment of this case is in tension with our knowledge that we should switch when confronted with the original Monty Hall problem.

Transcendental arguments become constitutive only when rational organization converges to a least admissible form from which every successful realization receives its structure. The meta-rational family is the clearest application domain for that claim. Polynomial presentation supplies a grammar of candidate package-worlds and the obligations they generate, but grammar alone is not deduction. -/- The constitutive upgrade is initiality. Once the structured admissibility doctrine attached to a semantically adequate polynomial presentation has an initial object, transcendental force becomes universal-property force: there is (...) a least admissible realization, and every other admissible realization uniquely receives its structure from it. The real burden of strong transcendental deduction is therefore an initiality claim. -/- The issue is not merely whether a rational package can stabilize, but whether there is a least admissible package-world from which all successful package-worlds receive their structure. Public-interface extraction, generic position-filling obligations, free repair, and re-entry yield a weak reconstruction of that family; in the finitary fragment, the present work shows both how initiality can be inherited from an ontological or evidential base and how free repair plus reflectivity can realize it at the package level. In one especially natural semantic reading, that inheritance runs through a reflective internal model presentation in an evidential topos; conversely, when package-level initiality is secured, it determines a least flattened ontological shadow in the restricted image of the doctrine. A public-repair obstruction theorem then shows how least package-worldhood can fail when admissible public completions branch, and a maximal-characterization theorem isolates the exact finitary boundary between the two. The resulting philosophical upshot is that rationality becomes constitutive only when its public burden collapses to a single governable visible condition; once the public frontier remains irreducibly plural, the doctrine yields admissible regimes rather than a least rational world. (shrink)

The standard Bayesian solution to the paradox of the ravens maintains that the degree of confirmation provided by seeing a nonblack nonraven is positive but negligible compared to that provided by seeing a black raven. I show that, unless we impose severe and unmotivated restrictions on the subject’s priors, this has the consequence that the cumulative confirmation provided by all the nonblack nonravens the subject expects to see is nonnegligible compared to the cumulative confirmation provided by all the black ravens (...) the subject expects to see. If this is so, however, then the paradox retains its full force. (shrink)

Formal theories of belief-credence interaction that satisfy the standard logical requirements on belief, such as conjunctive closure, face the problem of partition-sensitivity. According to these theories, a rational agent can believe X relative to one partitioning of possibilities, but the same belief may not be rational relative to some other partitioning, even when the agent’s evidence remains the same. Focusing on Leitgeb’s stability theory (Leitgeb, The stability of belief: How rational belief coheres with probability, Oxford University Press, 2017), which exemplifies (...) this problem, this paper aims to go beyond the simple partition-sensitivity of such theories. To address this and related problems, I connect the issues surrounding belief-credence interaction with Bayesian network theory and the notion of epistemic basing. The view that I defend and explicate is that rational belief corresponds to stably high probability within a network of interconnected propositions, where the network is structured by the relation of direct explanatory priority. As I argue, this view provides solutions to important problems regarding partition-sensitivity, specifically the problems involving irrelevant propositions and large partitions. In the final section, I extend the proposed theory and articulate the network-based notion of robustness of belief, which differentiates lottery-type beliefs from more ordinary beliefs. (shrink)

Bayes reappears across conceptually distant settings. We argue that this recurrence is neither accidental nor evidence that Bayes is the deepest universal law of rationality. Instead, we frame Bayes as the characteristic law of a recurrent public evidential interface. Rich evidential regimes must often pass into publicly accountable form by reflection, descent, quotient, or stabilized completion. When that happens, well-behaved public interfaces are fixed by subtopos modalities, polynomial functor structure fixes the admissible class of compositional update systems on the public (...) side, and Bayesian conditionalization is uniquely determined as the public law in the exact fixed-point case: the only update transformation that respects publicity, preserves relative support on the surviving fragment, normalizes, and composes under base change. -/- We organize this claim by four schemata: interface formation, Bayesian emergence, selection, and residuality. The first explains how a public evidential world is formed at all; the second explains why its exact law is Bayesian; the third explains why one interface or law is privileged among rivals; and the fourth explains what remains outside the public image. The payoff is both explanatory and classificatory: Bayes is exact on the public side, but the public side is generally non-faithful, so different cases can be read as different Bayes-defect types. The synthesis also adds a converse reconstruction problem: given a public Bayesian law, which richer upstream dynamics lift it, and how are those lifts classified by defect? (shrink)

Accuracy-first arguments show that Bayesian updating uniquely minimizes expected inaccuracy relative to a fixed, scoreable evidential representation. Accepting that theorem-level result, we ask a different question: what philosophical burden can such derivations bear? We argue that the strongest foundational reading of accuracy-first Bayesianism presupposes more than the mathematics itself establishes. Proper-scoring arguments require evidential content to inhabit an affine space on which inaccuracy is well defined, but diachronic coherence does not itself force affineness. Moreover, coherent global evidential order can be (...) non-affine when locally Bayesian evidential regimes fail to descend to a single global public structure. In such cases one may still recover a canonical affine shadow of the evidential order, and Bayes may be vindicated on that shadow. But the result is then conditional: it is a theorem about the affine approximation, not yet a full foundational vindication of the original evidential order. We therefore distinguish internal from foundational vindication, argue that accuracy-first succeeds at the former, and identify the further bridge condition under which vindication on an affine shadow can lift to the original evidential order. (shrink)

Scientific inquiry often supports disciplined Bayesian update only patchwise. Different instruments, evidential partitions, or reporting practices can each sustain a local public calculus without jointly yielding one global affine evidence space. The local/global problem for Bayesianism is therefore this: when do local canonical Bayes charts glue into one global canonical update law, and when does globalization fail by genuine obstruction rather than by mere local incoherence? -/- The core result is a local/global duality. In one direction, global canonical Bayes exists (...) exactly when local public quotients form effective descent data, that is, local quotient data that genuinely come from one common public object. In the other, a coherent but globally nonaffine update regime, not representable by one global affine public evidence space, can, under local canonicalizability conditions, be represented as a curved assemblage of locally Bayesian charts whose nonaffinity is measured by the associated descent obstruction. The obstruction therefore marks a limit on public epistemic integration, not a lapse into irrationality: inquiry can remain globally transport-coherent even when it admits no single global affine public evidence space. A standard-Borel Markov realization and several philosophical consequences place this geometry in the broader account of Bayesianness as affine evidence, now refined by the possibility of globally curved Bayesian order. (shrink)

Scientific inquiry often supports disciplined Bayesian update only patchwise. Different instruments, evidential partitions, or reporting practices can each sustain a local public calculus without jointly yielding one global affine evidence space. The local/global problem for Bayesianism is therefore this: when do local canonical Bayes charts glue into one global canonical update law, and when does globalization fail by genuine obstruction rather than by mere local incoherence? -/- The core result is a local/global duality. In one direction, global canonical Bayes exists (...) exactly when local public quotients form effective descent data, that is, local quotient data that genuinely come from one common public object. In the other, a coherent but globally nonaffine update regime, not representable by one global affine public evidence space, can, under local canonicalizability conditions, be represented as a curved assemblage of locally Bayesian charts whose nonaffinity is measured by the associated descent obstruction. The obstruction therefore marks a limit on public epistemic integration, not a lapse into irrationality: inquiry can remain globally transport-coherent even when it admits no single global affine public evidence space. A standard-Borel Markov realization and several philosophical consequences place this geometry in the broader account of Bayesianness as affine evidence, now refined by the possibility of globally curved Bayesian order. (shrink)

Theoreticity matters structurally, not merely taxonomically. This paper treats theoreticity as fiber-dependence relative to an observational projection: what is observational factors through the current public quotient, and what is theoretical still depends on the hidden fibers. Its central result identifies essential theoreticity with the maximal hidden invariant of an admissible architecture class, the largest family of hidden distinctions preserved across the allowed revision family. In non-canonical regimes, where observation is too weak to force a unique update law, that maximal hidden (...) invariant rigidifies inquiry by constraining admissible updates and revision paths. -/- Three further consequences follow. Essential hidden-role data can globalize from local evidential regimes when their fiber data descend coherently. In curved non-canonical settings, the same hidden invariant functions as a spatio-temporal stabilizer by deforming compatibly with local/global evidential geometry and preserving rectangular transport under the appropriate witness condition. Under stronger stabilized-completion hypotheses, it can also select a common admissible completion on which canonical observational order is restored and Bayesian update becomes recoverable again. The result is a structural account of theoreticity as hidden-role geometry and of essential hidden structure as the non-canonical stabilizer of disciplined inquiry. (shrink)

Explanatory adequacy descends to the public side of an admissible evidential architecture only under specific structural conditions. The core result is an explanatory descent criterion: explanatory structure is fully public exactly when the invariants required for explanatory adequacy factor through admissible public quotients. Public explanation is therefore a special success case rather than the default form of explanatory order. -/- Three main consequences follow. First, the descent criterion yields a minimal-support theory of explanation: explanatory adequacy depends not only on invariants, (...) but on the least hidden structure needed to preserve them. Second, this support logic induces a role-relative compressibility order: prediction tolerates the most compression, explanation less, and hidden-mediated Bayesian recovery less again. Third, failure of public descent yields both an anti-flattening trilemma and an Occam-disciplined account of Bayesian recovery: if hidden structure can restore Bayes on a designated fragment, there is a least hidden support sufficient to do so. A further operator-level result shows that even where term-level hiddenness is too coarse, explanatory or inferential machinery may remain essentially hidden. (shrink)

Reflective Bayesianization becomes the canonical update law of an observational regime exactly under specific structural conditions. The central result is an identification theorem: when the reflective quotient itself is observationally regular, unramified, and sufficiently separating, the update rule generated by disintegration is exactly reflective Bayesianization. In that setting, the Bayes report selected by equilibrium reasoning and the Bayes operator forced by strong observation are the same structure. -/- Two consequence branches follow. First, equilibrium semantics that previously lived only at the (...) quotient layer become semantics of the canonical operator itself, so Bayes/Nash commutation is internalized rather than merely transported. Second, once the resulting operator is strongly enough codable, the incompleteness ceiling proved for canonical observational update applies specifically to globally canonical Bayesianism. The philosophical payoff is that Bayesian update, equilibrium stabilization, and sufficiently strong observation are not rival foundations but convergent features of one public-score architecture. Bayes is neither a primitive law of rationality nor an arbitrary convention, but the coordinate calculus of a particular observational regime. (shrink)

In sufficiently expressive observational settings, sharpening observation into a globally unramified architecture forces a choice: one cannot preserve both conservativity of extension and decisive completeness of the observational calculus. We call this the Unramified observation–Conservativity–Observational decisiveness (UCO) trilemma. The key mechanism is that removing observational ramification turns updating on observation from a choice among admissible rules into a canonical internal operator. Once such an update calculus is internalized in an arithmetic-interpreting fragment, Gödel-Rosser incompleteness produces undecidable observational sentences; conservativity then reflects (...) this pressure back into the base theory, contradicting decisiveness. Philosophically, UCO sharpens Quine–Duhem holism and neo-Kantian views of framework change: resolving structural underdetermination often requires principled site enrichment, but global attempts to make observation fully lossless while preserving inherited commitments run into logical ceilings. (shrink)

Accuracy-first arguments derive Bayesian updating by assuming that evidential content is already representable as a scalar quantity evaluated by a proper scoring rule. This paper instead identifies structural conditions under which such scalarization is legitimate. We model evidence as interactive witness–challenge refinement and internalize updating in a Dialectica category over credal states. In this setting, feasibility of an update is equivalent to nonnegativity of a minimax security value, and diachronic coherence corresponds to compositional preservation of no-book status. Admissible policies admit (...) a canonical factorization into transport, interactive refinement, and optional selection. -/- Within this interactive universe we isolate the public-score fragment, the reflective subcategory of evidence types indistinguishable by strictly proper scoring rules. On this fragment, equilibrium semantics commutes with reflection: computing equilibria before or after collapse yields the same public reports. Bayesian updating appears exactly as the reporting equilibrium of the reflected evidence game. Outside this fragment the collapse is non-faithful: evidence types that coincide at the scalar level can remain strategically distinct in the interactive universe. A finite normal-form representation theorem and nonclassical LP stress tests make this boundary explicit. (shrink)

To model semantic choice in Bayesian updating under misspecification, we hold fixed a common update protocol—constraint refinement with envelope readout—together with a common selector and external proper scoring rule, and vary only semantic package geometry. Focusing on misspecification risk and semantic risk (with risk-functional choice treated as contextual), we prove minimal regime-dependent crossover theorems: classical semantics performs better when evidence constraints are truth-aligned, while non-classical packages can outperform in high-unreliability regimes with explicit thresholds in the proved minimal models. We operationalize (...) unreliability by widening admissible evidence bands (with directional bias), as one convenient proxy for model risk. The mechanisms are distinct: overlap slack in LP, deferral slack in intuitionistic semantics, and a propagation-gating quantity derived from relevant-logic implication structure. We also provide structural robustness arguments and an adaptive design sketch, while keeping full richer-family generalization as future work. The contribution is methodological: a disciplined interface for performance-based comparison of semantic commitments under model risk. (shrink)

Updating is not a single operation but a composite policy. This paper develops the Constraint--Transport Architecture (CTA), a modular framework that separates four roles: admissibility refinement, transport of prior states, optional posterior selection, and envelope projection to observable bid/ask prices. Rather than taking ratio conditioning as primitive, CTA isolates what coherence constrains and what requires additional normative commitments. It thus serves as a conditions-of-possibility framework, making explicit which parts of updating are forced by coherence and which are optional policy layers. (...) -/- We prove three core coherence-geometry results. First, soft constraint refinement is feasibility-exact under mild regularity: an evidential target is feasible exactly when it lies within prior envelope bounds. Second, posterior envelopes induced by refinement are coherent under standard bid/ask trading semantics. Third, nested refinement is diachronically safe against time books. -/- We then prove a conditional normal-form theorem: under affine score foliation and carrier compositionality, feasible update policies factor into context shift plus constraint enforcement, with optional selection. Relaxing affine foliation yields curved evidential geometries, showing that coherence alone does not privilege classical ratio rules. -/- Classical Bayesian and Jeffrey updating appear within CTA as specializations obtained by adding transport and selector commitments on top of refinement geometry. LP case studies serve as nonclassical stress tests, including contradiction-mass propagation and tight representability bounds in the finite two-atom setting. (shrink)

A unified meta-structural account synthesizes two tightly related strands of work on rational updating. The first strand (fixed-architecture decomposition) decomposes updating inside a fixed admissible architecture into distinct roles: admissibility refinement, state transport, optional posterior selection, and envelope projection to observable bid/ask prices. The second strand (drift/rectangularity analysis) studies diachronic rationality under evolving admissible architecture and identifies rectangularity/commutation as the feasibility condition for dynamic stability under representational drift. We show how to treat both strands as instances of a layered closure (...) regime: closure/interior operators on multiple representational layers whose commutation governs diachronic stability. -/- We formalize a three-layer framework (observable envelopes, credal geometry, and admissible architecture) and prove representation theorems showing when update policies admit a canonical factorization into orthogonal structural components. We then classify systematic extension modes: situations in which component-wise closure holds (each layer is locally coherent) but global factorization is non-canonical or commutation is obstructed, producing order effects, non-liftable observable policies, or time-book vulnerabilities under drift. Philosophically, this picture leads to a conditions-of possibility thesis: meta-rationality is governance of closure regimes that remain composable under self-iteration. (shrink)

Accuracy-first epistemology evaluates credences by scoring-rule inaccuracy and often uses a single scalar objective as the central criterion for inquiry within a representational scheme. This paper proves a no-go result under misspecification: an experiment can be epistemically preferred by the agent's own scalar accuracy criterion yet be worse by the same score at the true data-generating process. We isolate the mechanism with a score-divergence decomposition and a template proposition separating internal improvement from external truth-tracking. We prove a baseline binary log-score (...) reversal, a local robustness result under small true correlation, a Brier-score analogue, a finite-outcome log-score generalization, and a score-universal theorem at risk neutrality for all strictly proper scores on finite outcome spaces. For finite risk sensitivity, reversal attenuates but persists. The upshot is structural: within-model accuracy norms are indispensable but not self-sufficient for robust inquiry under model error, which requires additional norms governing representational adequacy and model criticism. (shrink)

Accuracy-first epistemology is often presented as if expected inaccuracy were the uniquely natural evaluation rule. This paper argues that this is a substantive normative choice: expectation is a risk-neutral aggregator of epistemic loss. We develop a risk-sensitive alternative that treats scoring-rule loss as an epistemic gamble and evaluates it with an entropic criterion. Philosophically, this reframes propriety-based sincerity as a special case, yields an endogenous distinction between private belief and public report, and models permissive disagreement as variation in epistemic risk (...) posture rather than a departure from accuracy-first methodology. -/- Technically, the framework delivers a unified structure linking cumulant-sensitive evaluation, a relative-entropy robustness dual, and tempered reporting operators, with an explicit map from private credence to risk-indexed report. We prove general existence and uniqueness results and show how variance appears as the first local correction to expectation. For log score we obtain a closed-form tempered reporting rule with clear limits from sincerity to maximal humility. For Brier score we prove strict convexity, characterize the optimizer by a Gibbs-type fixed point, derive an explicit local flattening direction, and establish the same maximal-humility limit. We also provide an axiomatic characterization of the escort family up to reparameterization, calibrate the canonical parameter map via the entropic log-score problem, and develop initial extensions to social aggregation, polarization, and diachronic consistency. (shrink)

Canonical Dutch-book arguments are usually presented in Boolean settings. Building on nonclassical convex-hull work and pragmatic-load cautions, we recast coherence as an extension problem: partial ticket prices are coherent when they extend to admissible valuations on gambles. We develop an order-enriched envelope interface that separates structural no-sure-loss constraints from semantic ticket design and from precisification assumptions. For LP, we prove both a finite-fragment representation theorem and a propositional global extension theorem (from finite coherence to global representing measures via compactness and (...) weak-* methods). We then derive LP-specific payoff consequences: the glut decomposition identity, coherence of "high-high" assignments, and a geometric result showing classical probabilism as an exposed face of LP credal geometry. The upshot is that probabilism is structural while complementarity is semantic: LP coherence supports both precise and imprecise regimes depending on ticket design and admissibility. (shrink)

The abortion debate has remained intractable for over fifty years because both sides impose binary thinking on continuous biological development. This paper argues that the impasse stems from a shared epistemic error, not from irreconcilable moral commitments. Whatever property one believes grounds moral status—consciousness, potentiality, human dignity, or future-like-ours—one faces irreducible uncertainty about when that property is present during fetal development. Bayesian epistemology shows that rational reasoning under such uncertainty requires graduated credences, and graduated credences require graduated protections. This framework (...) is neutral between first-order theories of moral status. A consciousness theorist who is uncertain when consciousness emerges, a potentiality theorist who is uncertain when potential becomes morally significant, and a dignity theorist who is uncertain when human dignity attaches all face the same structural problem—and the same structural solution. The paper demonstrates that both extreme positions (full status from conception, no status until birth) generate implications that their own advocates reject, revealing that both sides already implicitly track graduated rather than binary moral status. The Bayesian framework makes this implicit reasoning explicit and provides principled justification for what most people already believe: early and late abortion differ morally, even though we cannot identify a precise transformative moment. The achievement is not consensus on first-order questions but transformation of the debate's structure. Disagreement shifts from mutual accusations of moral blindness to legitimate epistemic disagreement about evidence weighting—a form of disagreement that permits productive engagement. (shrink)

The problem of evil is the most prominent argument against the existence of God. Skeptical theists contend that it is not a good argument. Their reasons for this contention vary widely, involving such notions as CORNEA (Condition Of ReasoNable Epistemic Access), epistemic appearances, ‘gratuitous’ evils, ‘levering’ evidence, and the representativeness of goods. This chapter aims to dispel some confusions about these notions, in particular by clarifying their roles within a probabilistic epistemology. In addition, the chapter develops new responses to the (...) problem of evil from both the phenomenal conception of evidence and the knowledge-first view of evidence. (shrink)

Strategic science skeptics criticize scientific claims solely to promote non-epistemic goals. I will analyze and debunk a philosophically neglected argument exploited by strategic science skeptics: the argument from disagreement. The core of this argument is that one should lower one’s confidence in a scientific claim when having learned that there is a scientific disagreement about this claim. I will develop a (Bayesian) Justificatory Account of Multiple Testimony to provide a normative characterization of how learning about agreements and disagreements is connected (...) to confirming and disconfirming scientific claims. I will use this account to debunk the argument from disagreement. (shrink)

I argue that a little learning is often dangerous even for ideal reasoners who are operating in extremely simple scenarios and know all the relevant facts about how the evidence is generated. More precisely, I show that, on many plausible ways of assigning value to a credence in a hypothesis H, ideal Bayesians should sometimes expect other ideal Bayesians to end up with a worse credence if they gather additional evidence, even when they agree completely about the likelihoods of the (...) evidence given both H and not-H. This provides a new reason for pessimism about the prospect of disagreeing individuals resolving their disagreement by consulting additional evidence. (shrink)

While many criteria of testability focus on the effect (or lack thereof) of observations on the theory, as suggested by Karl Popper, the more appropriate approach is to focus on the theory’s effect on observations, as suggested by A. J. Ayer and Elliott Sober. Under this assumption, Intelligent Design fails to be testable, and Creationism either is disconfirmed or, if it is shielded from disconfirmation by the modification of other theories, fails to be testable as well. Untestable claims can provide (...) a heuristic framework for the development of a testable theory, but to develop such a framework, Intelligent Design proponents would have to develop Intelligent Design by specifying the designer. (shrink)