Topics — What a work is about. Nucleation holds whatever reading the owner files here, whether or not an Anthology of the SOTA topic could also hold it; instruction for ML practice belongs to the anthology (ADR-048). The words are seeded from the clusters the reading-time triage of 2026-09-25 found, and grow by decision.
Epistemology — knowledge, belief and evidence — justification and rational credence, the value of knowledge, testimony, trust and expertise, and social and inductive-risk questions about evidence (group: philosophy).
60 of 1352 THEORY documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| THEORY-023 v4 | Current evidence cannot settle whether an AI system is conscious: mimicry undercuts behavioural evidence and architectural indicators presuppose the disputed computational functionalism, so the dispute is over what counts as evidence as well as the answer | Birch (2025), LIT-111: behavioural markers are “gamed”, and architectural markers are “Janus-faced”, giving a two-level deadlock. Schwitzgebel (LIT-191) argues mimicry undercuts but does not rebut, and that the functionalist theories face minimal instantiation. Seth (LIT-135) makes conscious AI conditional on functionalism plus substrate flexibility. Butlin et al. (LIT-056) assume functionalism. The readings agree the question is currently unsettleable. They contest whether that is permanent and which way the prior leans. This says nothing about whether any AI is conscious. Block (2002, LIT-639) argues that for a perfect functional duplicate on another substrate we lack even a conception of the evidence; Block (2009, LIT-632) states the biological premise from the other side. | Proposed |
| THEORY-053 | A treatment’s efficacy is not evidence for its theory of change: only change in the theorised process that is specific and comes before the outcome bears on it, and moving the process without moving the outcome counts against the theory | Salkovskis, Sighvatsson and Sigurdsson (LIT-538): “The effectiveness of a treatment is not evidence for the validity of the theory on which it is based”, with one exception, that changing the theory’s crucial factor without changing the problem disconfirms it. Hofmann and Hayes (LIT-544), from the other wing of CBT: a model whose process cannot be shown to apply “is incorrect and needs to be set aside” even if the intervention works, and a procedure that fails to move the process leaves the model untested. Active because it is close to a logical point and both state it. What it does not say is that mediation proves a mechanism; both sources deny that. | Active |
| THEORY-142 | Because the returns to exploitation are surer, nearer and closer to the decision than the returns to exploration, organizational learning refines exploitation faster and tends to become self-destructive in the long run; and where an organizational code learns only from members who deviate from it, slow socialization, a mix of slow and fast learners and moderate turnover raise what the organization knows, while without turnover a changing environment degrades its knowledge to chance | March (1991), LIT-729. Exploration’s uncertain, distant returns lose out to exploitation’s under learning from experience, trapping organizations in suboptimal equilibria. In a simulation of mutual learning between members and a code, deviation is the code’s only source of improvement, so fast socialization hurts the organization and moderate turnover helps it. Proposed: the dynamics are shown only in a stylized model, and the trade-off is argued, not measured. | Proposed |
| THEORY-143 v3 | When a group learns only about the options its members pursue and good evidence is scarce, sparser communication makes it more likely to settle on the better option, at a large cost in speed; when evidence is plentiful, sparsity only slows learning | Zollman (2007), LIT-724, simulated Bala and Goyal’s bandit model: cycles chose the better method more often than complete graphs, and more slowly; density, not centrality, predicted failure. Rosenstock et al. (2017), LIT-733, found the effect only where inquiry is hard: close alternatives, small groups, small batches. Proposed: all of it is simulation, and real communities’ place in parameter space is unknown. | Proposed |
| THEORY-146 v2 | In Holman and Bruner’s bandit model of honest Bayesian researchers whose beliefs funding cannot touch, who differ in method and whose influence and training of newcomers follow productivity, industry funding of those whose methods already favour its drug lowers, in the runs their figures report, the share of the community on the better treatment, and has no effect without methodological diversity; at the one setting reported, an independent funder that counts industry-funded trials toward its merit threshold lowers that share further, and one that excludes them raises it | Holman & Bruner (2017), LIT-744. In a Zollman bandit model with methodological diversity, productivity-weighted influence and turnover, industry funding of agents whose methods favour the drug lowers the chance of converging on the better action, though no agent’s beliefs respond to funding. A merit-based independent funder worsens it; one ignoring industry-funded trials counters it. Held as what the model shows; whether real fields work this way is QUESTION-315. | Active |
| THEORY-148 | A group learning by trying needs diversity in what its members pursue that lasts long enough to survive misleading early evidence and then gives way to consensus; limited communication and stubborn priors each supply it, and together they make it permanent, so the group never settles | Zollman (2010), LIT-735: in a bandit model with beta priors, sparse networks and extreme priors each maintain diversity of pursuit and each improve reliability, but combined they keep diversity alive so long that agents fail to converge. Rosenstock et al. LIT-733 find this the result that is robust across the models they test. Proposed: shown only in simulation. | Proposed |
| THEORY-151 v2 | Scientific objectivity belongs to a community’s critical practices rather than to individual method: because evidence bears on hypotheses only through background assumptions that no evidence supports, a community is objective to the degree that its criticism has venues, uptake, public standards and tempered equality of intellectual authority | Longino’s critical contextual empiricism, as she states it in the SEP (LIT-725). Underdetermination leaves background assumptions that only community criticism can check. Criticism confers objectivity only if it meets four norms: venues, uptake, public standards, and tempered equality. Proposed because the record reads the summary, not the books, and the norms’ epistemic benefit is argued rather than shown. | Proposed |
| THEORY-225 | No single-mechanism analysis of ordinary manipulation fits the clear cases: covert influence, covert intent, non-rational influence, trickery and pressure each misclassify a familiar tactic, so what unifies manipulation is unsettled | Noggle, SEP (2018, rev. 2026), LIT-1007, §§2.1–2.4, read in full in NOTE-744. A survey that argues for little; this is its one conclusion with moderate support, reached by giving each family of analyses a familiar tactic it misclassifies. The cases rest on shared intuitions about clear instances, not on argument. It leaves open common-factor and non-definitional accounts, and it is about ordinary manipulation, not the global manipulation of the free-will literature. | Proposed |
| THEORY-242 v3 | A spiral of silence can make a small, fixed minority opinion look far more widespread than it is without any member changing belief, and in the model that shows it the network’s direct distortion of what members see is small | Vilone and Polizzi (2024), LIT-939, read in full in NOTE-810. In agent-based models on scale-free networks with fixed private beliefs, a hard-core minority drives 55–60% of agents, nearly all of them in the majority, into silence, so that expressed and perceived opinion lean far toward the minority. The friendship-paradox distortion is small and of no fixed sign. This is simulation only. Beliefs never change, so nothing here shows a community coming to believe the minority view, and the majority is overturned only when the minority exceeds about 30%. | Active |
| THEORY-267 v3 | In Rasooly and Rozzi’s pre-registered randomised experiment on 817 Manifold play-money markets, about 59% of a ±5-point price shock that no trader’s information supported was still present after 60 days in unresolved markets, with about 25% undone in the first week and reversion slowing after it; their model of finitely many risk-averse traders gives such partial, slowing reversion even with no learning, and the data cannot tell that from learning | Rasooly and Rozzi (2025), LIT-958, read in full in NOTE-778. A pre-registered field experiment on 817 Manifold markets found about 59% of a random ±5-point shock still present after 60 days in unresolved markets. Reversion was 25% in week one and 6% more over the next three weeks. The paper’s model gets partial persistence from finite, risk-averse traders even with no learning. The data cannot say whether traders came to believe the shocked price. Its evidence that more traders or an outside estimate reduce persistence is correlational and weak. | Active |
| THEORY-273 v2 | Inductive and explanatory usefulness is necessary but not sufficient for a natural kind: it holds of classes that share any natural property, and on its own it cannot tell a natural kind from a conventional or pragmatic one | Bird and Tobin, SEP “Natural Kinds” (2008; rev. 2022), LIT-244, read in full in NOTE-769: §1.1.1’s criteria, §1.2.1 on induction over properties and relations, and §3’s report of Lemeire (2018). The necessity half is common ground; the insufficiency rests on counterexamples and, for the step to conventional kinds, on a reported argument. It does not say what does suffice, and it does not deny that usefulness is evidence. | Proposed |
| THEORY-290 v2 | In a two-debater, three-proposition model of deductive debate whose payoff trades true beliefs against changes of mind, the most truth-conducive equilibrium can fall short of the truth-optimal profile because one debater free-rides on the other’s belief-disrupting arguments, and in a narrow band of mixed equilibria a higher weight on truth gives lower equilibrium accuracy | Handfield, Garcia, Hilbe and Yeo (2025), LIT-966, read in full in NOTE-795. With payoff w·v − d (true beliefs against changes of mind), debate between two deductive reasoners is a cooperative dilemma at low and moderate truth weights. Typically one argues boldly and the other free-rides (a snowdrift), and at low weights both exit. A small dip in accuracy as w rises occurs in a narrow band of mixed equilibria. It is shown numerically for two players and three propositions. It concerns collective accuracy, not a belief the pair settles on, and at high truth weight in short debates there is no dilemma. The title claims the model only; groups and richer strategies are QUESTION-173. | Active |
| THEORY-295 v3 | In a community whose consensus model is replaced whenever a single sound pairwise model comparison favours the challenger, which models get proposed can hold the consensus on a false model while results reproduce almost perfectly, so reproducibility need not track convergence to truth | Devezer, Nardin, Baumgaertner and Buzbas (2019), LIT-988, read in full in NOTE-792. In a Markov-chain and agent-based model where each step tests one proposed linear model against the consensus by AIC or SC, replication cannot change the transition matrix. A boundary-testing population reproduces almost perfectly while rarely holding the true model, and correlations between reproducibility and truth depend on the strategy mix. The consensus is memoryless, scientists hold no beliefs, and an equal strategy mix is shown only never to be the worst, not best. | Active |
| THEORY-317 v3 | Epistemic injustice wrongs a person in her capacity as a knower, testimonially through a hearer’s prejudicially deflated credibility judgement, and hermeneutically through a gap in shared interpretive resources owed to her group’s relative powerlessness, with no hearer’s prejudice required | Fricker (2007), LIT-1004, known here only through her own three-page précis (Theoria 2008), read in NOTE-832, Skimmed. The two definitions and the claim that both are ethical wrongs, not only epistemic dysfunctions. It does not say that every gap in shared concepts is an injustice: on the précis’s terms the gap must be owed to the relative powerlessness of a social group the subject belongs to. | Proposed |
| THEORY-321 v4 | Under repeated DeGroot averaging in a strongly connected society, the consensus converges to the truth as the society grows if and only if the largest influence weight vanishes, so network structure alone, not the members’ common updating rule, decides whether the group belief is right | Golub & Jackson (2010), LIT-999, Proposition 2 with Lemma 1, Corollary 1, Proposition 3 and Theorem 1, read in NOTE-840 with the main proofs checked. The members are naïve, not sound: they ignore repeated information, and only the first round is Bayesian. The result is in a model with fixed weights, independent signals and n → ∞. Speed of consensus and its correctness are independent. | Active |
| THEORY-326 | Hermeneutical injustice is having a significant area of one’s social experience obscured from collective understanding owing to hermeneutical marginalisation, which may be systematic or incidental, so it does not require that the subject belong to a powerless social group | Fricker (2006), “Powerlessness and Social Interpretation” (LIT-1046, NOTE-859), drawn by her own note from the book’s final chapter, with her 2017 restatement (LIT-1048, NOTE-853). The generic definition covers a systematic case, where marginalisation is persistent and wide-ranging, and an incidental case, a “one-off moment of powerlessness” in a “highly localized patch” of experience, which a socially powerful man can suffer. It does not say that every conceptual gap is an injustice. A gap with no marginalisation behind it, like an undiagnosed illness, is “circumstantial epistemic bad luck”, and the obscured area must be of the subject’s social experience. | Proposed |
| THEORY-327 v2 | Fricker, in her 2003 paper and her 2017 retrospective, holds that testimonial injustice is an ingenuous misjudgement of a speaker’s credibility caused by prejudice, and that deliberate manipulation of credibility is a related but distinct wrong; and she accepts Anderson’s structural form, which is pre-emptive: a speaker’s word is never sought because of a structural prejudice, such as an un-refreshed list, that is no one’s fault | Fricker, in her own words, in two papers read in full: the paper that introduced testimonial injustice (2003, LIT-1049, NOTE-860), and her retrospective chapter (2017, LIT-1048, NOTE-853). A speaker receives the wrong degree of credibility “owing to a certain sort of unintended prejudice on the hearer’s part”. Deliberate manipulation of credibility is “a closely related but distinct injustice”. With Anderson, she accepts a structural form: when an un-refreshed list means some people’s views are never sought, “(through no one’s fault, let us add)”, that is pre-emptive testimonial injustice. It does not say every credibility deficit, or every omission from a list, is an injustice: a prejudice must cause it. | Active |
| THEORY-347 | Belief bases with the same logical closure can revise to opposite conclusions on the same input, so a closed belief set does not fix its own revision; partial meet contraction of a base need not satisfy Recovery, and bases keep distinct inconsistent states apart, while a belief-set model reproduces the base difference only by attaching a different selection mechanism to the same set | Hansson’s Stanford Encyclopedia survey (2006, revised 2026), LIT-1072, read in NOTE-866, with Alchourrón, Gärdenfors and Makinson (1985), LIT-1050, read in NOTE-874. Bases {p, q} and {p, p ↔ q} have one closure, yet revised by ¬p one yields q and the other ¬q: statically but not dynamically equivalent. Base contraction escapes the Recovery that AGM proves for closed theories, and many inconsistent bases exist where there is one inconsistent closed set. A base marks which beliefs are held in their own right and recomputes the rest; it is not a record of reasons, sources or defeated alternatives, and the survey gives no result that bases handle iterated revision better. | Active |
| THEORY-353 | For a logically closed theory, the contractions satisfying the six basic AGM postulates are exactly the partial meet contractions, and those satisfying the two supplementary postulates as well are exactly the ones whose selection of remainders is generated by a transitive relation; the theory alone does not fix the contraction, and the results concern one change of one fixed theory | Alchourrón, Gärdenfors and Makinson (1985), LIT-1050, read in NOTE-874, with Hansson’s survey, LIT-1072, read in NOTE-866. Contracting a closed theory A by x keeps what a selected set of maximal x-free subsets of A share; the six basic postulates characterize exactly these partial meet contractions (Observation 2.5), and with the two supplementary postulates exactly those selected by a transitive relation (Corollary 4.5). Revision follows by the Levi identity. A selection function or ordering must be added to A, and in the finite case distinct ones give distinct contractions. Nothing here concerns iterated change, where the selection comes from, or why a belief is held. | Active |
| THEORY-401 | A fair lottery on the natural numbers has a uniform, regular probability defined on every set of tickets, by normalised numerosity with values in the hyperrationals; it is fixed only up to a choice of free ultrafilter, it is hyperfinitely and not countably additive, and its standard part is ordinary asymptotic density | Wenmackers (2011), LIT-163, chapter 2, read in NOTE-139. P_num(A) = num(A)/α is defined on all of 𝒫(ℕ), gives each ticket 1/α > 0, and satisfies FAIR, ALL and a hypercountable form of SUM (eqs. 2.20–2.27). Proved, given the axiom of choice. It is not a free lunch: one solution per free ultrafilter, no invariance under relabelling, no countable additivity, and by the author’s own admission it is empirically indistinguishable from the finitely additive real-valued solution. | Active |
| THEORY-404 v2 | With a common prior, a sender who commits to a signal the receiver observes can induce any distribution of posteriors whose mean is the prior, so an openly designed experiment moves a fully informed Bayesian receiver whenever the concave closure of the sender’s value lies above its value at the prior; at the optimum, under the paper’s tie-breaking, the receiver is certain when she takes the sender’s worst action and indifferent when she takes another | Kamenica & Gentzkow (2011), LIT-1021, read in NOTE-717 with the proofs. Proved results (Proposition 1, Corollary 2, Propositions 4 and 5) for a single ideal Bayesian receiver, a finite state space, commitment and sender-preferred tie-breaking. A judge who knows the investigation was designed to convict convicts 60 percent of defendants when 70 percent are innocent. The paper does not analyse manipulation, does not compute the receiver’s welfare and does not cite Blackwell. | Active |
| THEORY-415 | Nothing has all four marks that qualia are said to have, being ineffable, intrinsic, private and directly apprehensible, because neither the subject nor an outside tester can tell a change in a quale from a change in the subject’s reactions to it; what is real in their place is the practical ineffability and idiosyncrasy of information-rich property detection | Dennett (1988), LIT-433, read in full in NOTE-357. Argued by fifteen intuition pumps, not by proof, and the author says so. The central step rests on a stipulation that two surgeries, or two adaptation stories, produce identical introspective effects. NOTE-357 filed no THEORY because the record does not hold the replies (Shoemaker 1981, Block 1980); this account states the claim as the reading supports it and leaves it Proposed for that reason. Its “there simply are no qualia at all” is a tactical choice of words by the author’s endnote 2. | Proposed |
| THEORY-422 | In the standard formal models of group learning, no communication network is best for every epistemic problem: costly transmission favours minimally connected or star networks, pooling of estimates by averaging favours regular networks, and learning by trying in bandit problems can favour sparse ones, so a group’s network cannot be judged until its learning problem is known | Zollman (2013), LIT-716, a survey read in NOTE-567. Each result it gathers is proved or simulated in its source (Bala and Goyal, Jackson and Wolinsky, DeGroot, Hegselmann and Krause, Zollman). Under their idealisations the optimal structure changes with the problem, so neither “too little communication” nor unequal centrality is by itself an epistemic pathology. It is a claim about models: utilitarian optimality only, with fairness, robustness and the route to the optimum left out. | Active |
| THEORY-433 | Whether a machine thinks is best replaced by whether a digital computer can play the imitation game as well as a man, because the reasons for withholding thought from a machine that does would equally withhold it from other people, whose thinking we also infer from what they say; the objections from consciousness, mathematical limits and informality of behaviour fail on that test or rest on unproved premises | Turing (1950), LIT-394, read in NOTE-355. The imitation game is offered in place of a definition of thinking, not as a test of it, and Turing says he has “no very convincing arguments of a positive nature”. The account’s force is the other-minds parity of §6.4, which assumes, as §2 states and does not argue, that the machine’s best strategy is to answer as a man would; a machine built to imitate human output is the case that assumption does not cover. It makes no claim about consciousness. Proposed, and a rival of Block’s psychologism. | Proposed |
| THEORY-438 | The understanding deep networks reach is characteristically fractured: it can be real, by the test of stable tracking under structure-preserving perturbation, but it is symbolically misaligned, non-reductive, locally fractured, proxy-dependent and representationally fragmented, because the loss rewards any predictive patchwork and does not press it to unify | Freeborn (2026), LIT-131, read in NOTE-123. The Fractured Understanding Hypothesis is argued from four small demonstrations and cited literature, with no test of how common the five features are. The abstract names three features and §7.5 five. The paper also applies its own criteria inconsistently: its Othello verdict uses post hoc probes that it excludes from the agent. The definition of understanding it rests on is an explication, not a finding. | Proposed |
| THEORY-442 | Where the non-epistemic consequences of error are foreseeable, sound scientific reasoning requires weighing them in the internal choices that carry inductive risk, and non-epistemic values then enter legitimately only indirectly, through the costs of error, never as reasons to think a claim true | Douglas (2000), LIT-139, read in NOTE-145. The normative half of the paper. It rests on a premise the paper declares “beyond the scope” and does not argue, that scientists are morally responsible for foreseeable consequences of error (p. 563). The direct/indirect line is drawn sharply in the conclusion but only sketched, and the reply that scientists could report uncertainty instead of accepting is not engaged. | Proposed |
| THEORY-451 | A singular causal claim, that C will contribute or did contribute to E in a particular setting, is warranted only by joint evidence on each step of the causal process there, its causal principle, its support factors and the absence of its derailers, which no single method supplies; a well-run RCT secures at most the study population’s average effect, and often not even that rigorously, and its result is not by its method evidence about any other population | Cartwright (2021), LIT-213, read in NOTE-161. The point that RCT results are not by the method evidence about another population (save under random sampling from it) is strong and standard. That no method or handful of methods can warrant a singular claim is argued from the diversity of facts a causal-process theory of change needs, with one illustrative case. That such theories yield reliable predictions is asserted, with no track record; the paper gives no method for weighing the evidence and little on transport. The critique of RCT rigour is internal to the author’s own reconstruction of “rigour”. | Proposed |
| THEORY-491 | Of the five standard arguments that randomisation gives a clinical trial special evidential weight, only the selection-bias argument is sound, and it supports only taking allocation out of the experimenters’ hands; given balance on plausible confounders, a placebo or active control and blinding, randomisation is argued to add nothing, since balance on unknown confounders is not even probable in the one division a trial makes | Worrall (2007), LIT-175, read in NOTE-126. The case against the unknown-confounders argument is conceptual: frequentist balance is a long-run claim about repeated divisions, and a trial divides once. It does not show that randomised designs are no more reliable in practice than matched ones, and the meta-research cited is mixed. The selection-bias argument is conceded as “cast-iron”, so the account does not say randomisation is worthless. Haphazard allocation is asserted to serve as well, without discussing whether such rules are predictable. | Proposed |
| THEORY-506 | Every well-developed theory of how mind fits into the cosmos is bizarre, contrary to what most people confidently believe, and none is epistemically compelling, because common sense about the metaphysics of mind is incoherent; so no coarse position, materialism included, deserves credence much above 50% | Schwitzgebel (2024), LIT-216, read in full from the June 2022 final manuscript in NOTE-188; the published body text was not reached. Chapter 3’s Universal Bizarreness and Universal Dubiety, with the other chapters as case studies. Bizarreness is an abductive generalization over the theories actually developed, with a standing challenge to find a counterexample, not a result about all possible theories. Dubiety rests on expert disagreement, the lack of any method and cosmological uncertainty, and the 50% figure is the author’s judgment. The author’s own credences lean materialist, at about 55%. | Proposed |
| THEORY-537 | Warrant for a type-level causal claim in biomedical and social science comes from eliminating the relevant alternative accounts of its support, not from the kind of study that produced the data, so a randomised trial earns its standing only by eliminating many alternatives at once | Reiss (2015), LIT-118, read in NOTE-197 from the 2015 Durham working paper; the refereed version was not read. A conceptual proposal with one historical illustration, the 1950s smoking and lung cancer controversy, which the author says is not a test. What counts as relevant, and when to stop probing, is handed to context, and the four grades of warrant are admitted to be arbitrary. | Proposed |
| THEORY-550 | Unsupervised clustering is not made epistemically superior to supervised classification by the absence of labels: it is sensitive to sample bias, highly sensitive to how inputs are represented, and overfits, with the number of clusters acting as a capacity control just as parametric complexity does in supervised learning | Søgaard (2024), LIT-162, read in NOTE-108. A rebuttal of two arguments for clustering’s superiority, the No-Bias Argument and the Simplicity-Truth Argument, by toy counterexample and by the technical literature on overfitting in unsupervised learning (Hansen & Larsen 1996). The negative claims are firm. The paper’s positive step, that supervised classification is “at least as” justified, is under-argued and is not part of this account. | Active |
| THEORY-556 | Clustering, abstraction and generative modelling can be how we come to identify natural kinds, essential properties and unrealized possibilities, but kinds, essences and possibilities are not by definition what such algorithms would find under ideal conditions, because nothing guarantees they are effectively computable; so an algorithm’s output is evidence about a kind, neither objective nor vacuous, and not what makes it one | Watson (2023), LIT-187, read in full in NOTE-172. Three pairs of theses, one per task: the epistemic one (EC, EA, EG) accepted, the ontological one (OC, OA, OG) rejected, each on the same computability ground. The rejection is a bare possibility argument with no example of an uncomputable kind, and the acceptance is hedged (“or something very much like them”) so far that it is hard to falsify. The pragmatic, context-relative view of kinds is asserted, not defended. Self-supervised learning and foundation models are outside its scope. | Proposed |
| THEORY-569 | Measured as the contrast between a set of explanantia’s CohOG+ coherence with the explanandum and with its negation, explanatory power satisfies Roche and Sober’s Two Likelihood Inequalities for single hypotheses and fails the One, and in one causal chain and one collider it ranks the richer set above the sparser over a range of priors, where no two-place TLI measure can rank a conjunction so | Hartmann & Trpin (2026), LIT-165, read in NOTE-168, which recomputed all three numerical claims. One theorem (TLI for singleton explanans sets), one counterexample (OLI fails) and two parameterised existence examples (a distal cause for P(H1) < .7; a negatively interacting cause for P(H1) = P(H2) < .75). The escape from Roche & Sober is representational: a set is scored differently from the conjunction, and the two-place theorems still hold for conjunctions. No characterisation says when the richer set wins, and the paper’s claim that the measure “accommodates key scientific judgments” is more than the body shows. | Active |
| THEORY-572 | Where an effect is not yet well conceptualised, replication cannot say which effect exists: direct replication bears on random error, conceptual replication presupposes that two operationalisations pick out the same effect, and an effect is delineated instead by probing for systematic error | Feest (2019), LIT-161, read in NOTE-196 from the PSA 2018 manuscript; the version of record was not read. An informal argument with one historical case, the Mozart effect, declared “typical” without further evidence. The claim that direct replication says “nothing” about systematic error is overstated: multi-lab replications vary some tacit lab-level factors. The conclusion is scoped by the author to effects not yet well conceptualised, and to internal validity; the title and abstract drop that hedge. | Proposed |
| THEORY-577 | Our judgments that consciousness poses a hard problem are behaviour, so explaining them is an easy problem that can be solved without mentioning consciousness; and a theory that names a mechanism as the basis of consciousness owes an account of how that mechanism produces those judgments, which IIT and biological theories struggle to give, since a functional duplicate with zero integrated information, or in silicon, would make the same reports | Chalmers (2018), LIT-425, read in full in NOTE-354. The meta-problem follows from the physical explicability of behaviour; the challenge is a constraint argued from it, “not a knockdown argument”. It assumes that a zero-Φ simulation has the same intuitions on a functional reading, which an IIT theorist can deny. The paper’s own favoured solution, opaque introspective models attributing primitive relations of acquaintance, is “closer to a fleshed-out explanandum” than an explanation, and the data the programme needs mostly do not exist. | Proposed |
| THEORY-596 | To ascribe beliefs and desires is to predict a system by assuming it is rational and its beliefs mostly true, so a wholly false belief system has no coherent intentional description and contradictory beliefs no stable one, and the believer, like the person built on that notion, is an idealisation that real creatures only approximate | Dennett (1971), LIT-440, read in NOTE-350, and his reply to Stich (1980), LIT-456, read in NOTE-371. Belief ascription presupposes optimal design relative to goals, so “a false belief system is a conceptual impossibility” and, following Quine, the evidence for ascribing p is the best evidence against ascribing a contradictory q. In 1980 Dennett concedes the price: the intentional system, and the person, is “an unrealizable idealization”. Argued from the logic of ascription and from evolution by assertion; no data. It is a claim about what ascription requires, not about what is in heads. | Proposed |
| THEORY-601 | Among US undergraduates in the HERI surveys, students with higher verbal ability and stronger self-reported intellectual dispositions select into philosophy, and after adjusting for their entering SAT or freshman scores philosophy majors still score higher than other majors on GRE Verbal (SMD 0.48) and LSAT (0.35) but not on GRE Quantitative, and report more Habits of Mind and Pluralistic Orientation; the adjusted advantage is an association, not a shown effect of studying philosophy | Prinzing & Vazquez (2025), LIT-155, read in full in NOTE-167. HERI/CIRP freshman–senior data on 649,511 students at 804 institutions, 4,843 philosophy majors. Selection is real (SAT Verbal OR 1.57, SAT Math none). Baseline-adjusted mixed models and an IPTW check agree on the verbal and logical-test advantage and on its absence for quantitative tests. All test scores are self-reported, the SAT is an imperfect baseline for the LSAT, and nothing rules out faster verbal growth or harder test preparation among those who choose philosophy, so the paper’s causal title is not part of this account. | Active |
| THEORY-610 | What survives radical theory change in a mature science, and so what its predictive success warrants belief in, is mathematical structure, equations carried over intact or recovered as limiting cases, and not the ontology that interpreted them, as Fresnel’s reflection and refraction equations survive into Maxwell’s theory while the elastic-solid ether does not | Worrall (1989), LIT-315, read in NOTE-297. Epistemic structural realism, credited to Poincaré, offered as the one position that keeps both the no-miracles argument and the pessimistic induction. The support is one worked case, which Worrall himself calls unrepresentative because the equations survive intact, and an appeal to the correspondence principle with one further example. “Structure” and “structural approximation” are undefined. The view is agnostic about what there is; the ontic turn, Ramsey sentences and the Newman objection belong to later work. | Proposed |
| THEORY-615 | Stubborn distrust of scientific experts can follow validly from the truster’s position: one with little say in collective decisions, who has reason to think the experts’ service ethos leaves out their core interests, has a reason to distrust an expert because of the expert’s prestige, whatever the expert’s competence or goodwill | Desmond (2022), LIT-158, read in NOTE-113. A “how-possibly” argument, not a demonstration: a five-step scheme (§6, pp. 16–17) from status vulnerability and service doubt to distrust “in virtue of” prestige. The paper says it shows validity only. It leaves open whether the premises ever hold in a liberal democracy, gives no data linking status to distrust, and does not compare its account with testimonial injustice or the empirical predictors of conspiracy belief. The abstract’s “justifiably” claims more than the body. | Proposed |
| THEORY-633 | Experiences have no phenomenal properties: introspection represents physical, quasi-phenomenal properties of sensory states as phenomenal, so what needs explaining is why experiences seem to have phenomenal properties, not how a brain produces them; a weaker view that keeps a deflated phenomenal residue is unstable unless it can say what that residue is | Frankish (2016), LIT-458, read in NOTE-384, names the view and the “illusion problem”; Dennett’s reply to Chalmers (1996), LIT-397 (NOTE-346), gives its earlier form, that subtracting every function from one’s consciousness leaves only a conviction; Frankish (2025), LIT-103 (NOTE-143), applies it to valence. A research programme, not a theory: Frankish proposes no mechanism, the instability of weak illusionism is argued by challenge, the case is abductive, and Dennett’s subtraction is an introspective assertion. It keeps consciousness in an inclusive, functional sense and denies only phenomenal properties. | Proposed |
| THEORY-637 | Treating openness in science as the unlimited sharing of research objects, with transparency ranked before reproducibility and inclusion, can narrow the epistemic diversity of research and entrench epistemic injustice, and this lowers the reliability of what is learned, so sharing is neither a necessary starting point nor a sufficient condition for reliable open science | Leonelli (2023), LIT-189, read in NOTE-098. An openly normative argument from four illustrative cases (GISAID, software stigma, crop-data linkage, the reproducibility debate) and the author’s earlier qualitative work. That sharing-first open science can reduce diversity and worsen injustice is moderately supported; that this degrades reliability is argued, not measured. The positive proposal, openness as judicious connection, gives no decision rule, and the critique targets an ideal type the author admits many initiatives do not fit. | Proposed |
| THEORY-639 | Inductive risk arises not only when a hypothesis is accepted but in choosing a significance level, in characterising data and in interpreting results: in the dioxin cancer studies each of these choices traded false positives against false negatives, three panels scored the same rat-liver slides differently, and threshold and no-threshold readings of the same data underlie regulatory limits about 100 times apart | Douglas (2000), LIT-139, read in NOTE-145. Three worked cases from one domain, dioxin toxicology and its regulation, on secondary data (EPA 1994; Kociba et al. 1978; Goodman & Sauer 1992). The structural point and the variation in slide scoring are strongly supported. The 100-fold figure is cited from Finkel (1988) and Greenlee et al. (1991). This is the descriptive half of the paper; the claim that values are therefore required is THEORY-442. | Active |
| THEORY-647 v2 | If outright belief is credence ultraclose to 1 relative to a context level, conjunction holds for any standard (‘few’) number of believed propositions and fails for an ultralarge number, which dissolves the Lottery Paradox as an accumulation of negligible rounding errors | Wenmackers (2011), LIT-163, chapter 3, read in NOTE-139. Stratified Belief, B(x) iff P(x) ≃_V 1 in Hrbacek’s relative analysis, satisfies a Stratified Conjunction Principle for a standard number of conjuncts (proved) and fails for an ultralarge number. The proofs hold inside the framework. That belief works this way is argued: any explicit threshold collapses the model back to the Lockean one, the author calls it “crude”, and knowledge is set aside. | Proposed |
| THEORY-649 | When agents holding uniformly random consistent theories over two or three atomic sentences each average their theories once, bit by bit, with all peers within a Hamming bound of confidence, an individual agent ends up with the inconsistent theory at most 1.81% of the time, but the chance that some agent in the group does reaches 6.4% for two sentences and 16.9% for three | Wenmackers (2011), LIT-163, chapter 5, read in NOTE-139. An exact count, extended by sampling, of how often one round of bounded-confidence averaging over theories yields the inconsistent theory. The per-agent fraction F_AG peaks at 1.8112% (M = 2) and 0.3215% (M = 3); the per-profile fraction F_OP reaches 6.4298% and 16.867%. The thesis’s headline, that the risk “can be made arbitrarily small” by adding sentences, rests on two values of M and is contradicted for F_OP. | Active |
| THEORY-810 | Prediction-market prices keep much of an arbitrary price shock that no trader’s information supports for weeks, rather than correcting it within a short transition phase | The general reading that THEORY-267 held before the promotion audit narrowed it to the 817 Manifold play-money markets measured. Rasooly and Rozzi (2025), LIT-958, read in NOTE-778, set the result against the received view (Wolfers and Zitzewitz, 2004) that manipulation matters only briefly, and offer it as a finding about prediction markets. Their evidence is one platform, at play-money stakes. Filed under ADR-047 v4 so the wide form stays a thesis evidence can confirm or reject. | Proposed |
| THEORY-814 | In scientific communities that settle on a consensus model by sound model comparisons, high reproducibility of results is not evidence that the consensus is true, because which alternatives get proposed can hold a false consensus whose results reproduce almost perfectly | The general reading that THEORY-295 held before the promotion audit narrowed it to the memoryless consensus model it proves. Devezer, Nardin, Baumgaertner and Buzbas (2019), LIT-988, read in NOTE-792, present the decoupling of reproducibility from truth as a lesson for science, not only for their model. Their model has no beliefs, no learning and no memory in the consensus. Filed under ADR-047 v4 so the wide form stays a thesis evidence can confirm or reject. | Proposed |
| THEORY-829 | In Hrbacek’s relative analysis, if belief at a context level is credence ultraclose to 1 at that level, the conjunction of any standard number of believed propositions is believed, while for an ultralarge number it can fail: in a fair lottery with an ultralarge number of tickets each ‘this ticket loses’ is believed and their conjunction has probability 0 | Wenmackers (2011), LIT-163, Ch. 3 (Stratified Belief, Def. 1; the Stratified Conjunction Principle, Def. 2, eq. 3.5 and §3.6.3.2), read in full in NOTE-139; re-derived by the promotion audit from finite subadditivity and Hrbacek’s Closure principle. A result inside the framework: it says nothing about whether belief has levels, and any explicit level boundary collapses the model back to a fixed threshold. | Active |
| THEORY-834 v2 | A member of a discipline who has been trained into its paradigms yet remains rooted in experience those paradigms omit or distort is better placed than its insiders to perceive those omissions and distortions as anomalies, and better placed than outsiders who share the experience to have them taken up, because insiders share the assumptions that produced the anomalies and outsiders lack the professional authority to press them | Collins (1986), LIT-1312, read in an authorized Portuguese translation in NOTE-1042: the “outsider within”. An argument from the sociology of knowledge and Kuhn, illustrated by anomalies Black feminist sociologists did detect. It does not say that every outsider within sees them, that insiders cannot, or that what the outsider within reports is true; and it offers no comparison of who detects what. | Proposed |
| THEORY-845 | In Weatherall, O’Connor and Bruner’s network model a policy maker’s credence depends on the evidence it has received only through the total excess of successes over failures of the new action, so a propagandist’s effect is the excess of failures it adds; for the same pulls, running single-trial studies and publishing the failures adds at least as large an excess as any other division into studies, which is why biased production does best with the smallest studies | Weatherall, O’Connor and Bruner (2020; arXiv 2018), LIT-1341, read in NOTE-1038. With two symmetric hypotheses, each result’s likelihood ratio depends only on successes minus failures, so the audience’s belief is a monotone function of the cumulative excess; whoever chooses what the audience sees chooses that number. It holds for the paper’s two-point hypothesis space and naive updating, and says nothing about richer hypothesis spaces or audiences that model the selection. | Active |
| THEORY-855 | In Bala–Goyal network models of unbiased scientists testing a slightly better new action, a propagandist who forwards to Bayesian policy makers only those real results that favour the worse action can leave the policy makers near-certain of the false theory while the scientists converge on the true one; the effect weakens as the policy makers hear more scientists, as the better action’s advantage grows and as each study gets larger | Weatherall, O’Connor and Bruner (2020; arXiv 2018), LIT-1341, read in NOTE-1038. No result is fabricated and no scientist is biased; the audience is misled because it updates on a selected sample as if it were not selected. It is a property of a stylized model, which the authors call “how-potentially”; it does not show that any real audience was misled this way, nor that an audience aware of the selection would be. | Active |
| THEORY-857 | Remedies for algorithmic governance offered as access or as transparency can run against the people they are meant to serve: inclusion in a data system also gives its operators access to the included, and transparency can be demanded of the governed while the governing system stays opaque, so what an access or transparency remedy does depends on who is made visible to whom | James, Hynes, Whelan, Dreher and Humphry (2023), LIT-1342, read in NOTE-1041. “Access is however bidirectional, and access ‘down’ for powerful players is politically consequential”; under Robo-debt transparency “was arguably demanded of welfare recipients”. The paper does not show that such remedies usually backfire, or say when they will. It also does not claim that transparency is worthless. | Proposed |
| THEORY-864 | When a generative model reproduces a prejudiced narrative learned from its training data, the resulting testimonial injustice runs through a credibility excess assigned to the model, so that hearers believe the amplified account over the testimony of those it discredits, rather than through a hearer’s prejudiced judgement of a present speaker | Kay, Kasirzadeh and Mohamed (2024), LIT-1338, read in NOTE-1049: “In the algorithmic setting, the injustice requires a credibility excess assigned to the algorithm”. A conceptual analysis, illustrated with reported cases. It does not say that users do overtrust models (the paper measures nothing), that deliberate fabrication is testimonial injustice in Fricker’s sense, or that the amplification is a loop with any stated condition. | Proposed |
| THEORY-877 | Learning a generative model of sensory data extracts real patterns in Dennett’s sense, so the manifest image is a structural description of the sensory signal, constrained by its external source but silent on what realizes the structure; two subjects whose sensory streams match, one envatted, therefore have equally veridical manifest images, and what differs between them, the realizers, lies outside what perception can represent | Gładziejewski (2025), LIT-1360, read in full in NOTE-1056, §§2–3, drawing on Chalmers’s structuralism and veridicalism (2018, 2022), which the record does not hold. The cognitive half rests on a realist reading of predictive processing and on an analogy between free energy and description length. The veridicalism follows only if manifest content is wholly structural, which is assumed. Projective properties (cuteness) are allowed at one end of a continuum. | Proposed |
| THEORY-884 v2 | A retained state supports exact answers to a declared family of questions after every finite sequence of a declared family of total deterministic revisions exactly when each of its fibres lies within one class of the coarsest revision-invariant refinement of agreement on current answers; that refinement can be strictly finer than what present answers require, and at most K* − K revision sequences, beyond the current answers, are needed to tell its K* classes apart | Aevyra (2026), The Ledger of Will, LIT-1376, read in full in NOTE-1091: Proposition 4, “revision closure”, with its proof in §4.4 and finite checks in Appendix A.9. The authors call it “a finite-machine refinement construction in the tradition of Moore and Nerode”, not a new theorem. It answers QUESTION-112 in one case only: a finite family of total deterministic revisions declared in advance. It does not say how to choose the family, which is that question’s hard part, and it is not an efficient construction. | Active |
| THEORY-943 v2 | When a center with no ground truth pays each rater a strictly proper scoring rule on the distribution of a reference rater’s report implied by her own report, under a common prior and a known signal model in which raters’ signals are stochastically relevant to one another, truthful reporting is a strict Nash equilibrium, but not necessarily the only one: in the authors’ two-signal example under the log rule, every rater always reporting h, or always reporting l, is also an equilibrium | Miller, Resnick and Zeckhauser’s Proposition 1, read in full in NOTE-1139 from the January 2005 draft, with the proof and the worked example (p(H) = .5, f(h | H) = .85, f(h | L) = .45; honest l scores −0.69, false h −0.76 under the log rule) re-derived. The same example has two uninformative equilibria, which the authors set aside as not focal. The scheme replaces ground truth with the center’s model of how honest signals correlate. | Active |
| THEORY-1072 | In Simon’s analysis, whether a problem is well structured is a relation between the problem and a bounded solver, judged by practicable rather than in-principle computability, so it comes in degrees and has no sharp boundary; on that standard even theorem proving and chess fail some of the conditions, chess is well structured in the small and ill structured in the large, and problems presented by the world become well structured only in being prepared for a solver | Simon (1973), LIT-1634, read in NOTE-1232 from a scan of the published article. Active as a conceptual argument the paper makes in full: six conditions for a well structured problem, read with condition 6 as practicable computation, worked through for theorem proving and chess. It does not say that structuring is the whole of problem solving, which Simon rejects, or that any problem is ill structured for every solver. | Active |
| THEORY-1084 v2 | Simon proposes that ill structured problems can be handled by an alternation between a solver working on a problem that is well structured at each moment and a recognition and retrieval system that keeps modifying the problem space from memory or the environment, that a serial solver gains nothing by assembling the problem space at the outset, and that nothing but the size of the knowledge base may separate ill from well structured problems, so that no qualitatively new mechanisms are needed | Simon (1973), LIT-1634, read in NOTE-1232. The alternation is described for an architect, matched to a 1923 account of battleship design by an organization and to the NSS chess program, and argued from the seriality of the solver; nothing is implemented or tested, and Simon states the last part as what his analysis “suggests” and as “reason to be optimistic”. Active as a report of the proposal; whether it is right, which wants a test, is THEORY-1216, Proposed. | Active |
| THEORY-1173 | Testimonial injustice is an ingenuous misjudgement of a speaker’s credibility caused by prejudice, not a deliberate manipulation of it, and it can be structural and pre-emptive, when a speaker’s word is never sought because of a structural prejudice that is no one’s fault | The wide form THEORY-327 held before the twelfth status review narrowed it, kept as a Proposed reading by the ADR-047 kind-1 sweep of 2026-10-30: the source asserts it, and THEORY-327, Active, is its evidence. Fricker states the account as what testimonial injustice is, not only as her usage; the record holds it as hers, in two papers read in full. Whether the book states it so, and whether it is right about the wrong, is open. | Proposed |
| THEORY-1174 | Industry can bias a scientific community toward a worse treatment without corrupting any individual, by selectively funding researchers whose methods already favour its product, provided researchers differ in method and both influence and the training of newcomers follow productivity; an independent funder that discounts industry-funded work counteracts this, and one that funds on productivity alone makes it worse | The wide form THEORY-146 held before the eleventh status review narrowed it, kept as a Proposed reading by the ADR-047 kind-1 sweep of 2026-10-30: the source asserts it, and THEORY-146, Active, is its evidence. The old title’s “counteracts this only if it discounts” is restated as “that discounts … counteracts this”: the paper compares two funder policies and asserts no necessity, a mis-statement left with THEORY-146’s note. | Proposed |
| THEORY-1216 | Ill structured problems can be handled by an alternation between a solver working on a problem that is well structured at each moment and a recognition and retrieval system that keeps modifying the problem space from memory or the environment, a serial solver gains nothing by assembling the problem space at the outset, and nothing but the size of the knowledge base may separate ill from well structured problems, so that no qualitatively new mechanisms are needed | The thesis Simon (1973), LIT-1634, proposes, kept here when the twenty-second status review promoted THEORY-1084 as a report of the proposal (ADR-065; ADR-047, kind 1). The paper describes the architecture, matches it to an architect, to battleship design and to a chess program, and argues from the seriality of the solver; nothing is built or tested, and Simon states the last part as what his analysis “suggests”. | Proposed |