{"path":"research/moral-convergence-metaethics-and-psychology.md","content":"# The Convergence Wager: Metaethics and Moral Psychology\n\n**Date:** 2026-07-05\n**Status:** AI-conducted research (part of the project's research-agent corpus). Sources were gathered by web research against primary academic literature; conclusions are the synthesizing agent's own and should be read as input to deliberation, not as settled doctrine.\n\n---\n\n## 1. Framing: what exactly is being wagered?\n\nDeliberus rests on a wager (explored in the companion documents *Convergence*, *Depth*, and *Bridging*): that most deep disagreement among people of good will is semantic confusion plus differently-interpreted feasibility sitting on top of broadly shared care — and that even value *weightings* (liberty vs. safety, near vs. far) decompose into constituent parts, so that with enough decomposition, views converge. A second, companion claim: we can resolve which views are better-informed and better-reasoned than others.\n\nThis document assesses that wager against metaethics and empirical moral psychology. To make the assessment tractable, the wager decomposes — fittingly — into four claims:\n\n- **C1 (semantic):** a large share of apparent deep disagreement is verbal — different concepts travelling under the same words.\n- **C2 (factual/instrumental):** much of the remainder tracks disagreements about facts, consequences, and feasibility rather than terminal values.\n- **C3 (weighting-decomposition):** residual differences in value *weightings* themselves decompose, and under sufficient decomposition plus idealized reasoning, converge.\n- **C4 (epistemic asymmetry):** independent of convergence, some positions can be shown to be better-informed and better-reasoned than others.\n\nThe claims have very different evidential standing. Headline finding: **C1, C2, and C4 are well-supported; C3 is a genuinely open philosophical bet — expert opinion leans against its strong version while supporting a substantial weak version.** The distinction matters for what the platform should promise versus what it should measure.\n\n---\n\n## 2. The convergentism debate in metaethics\n\n### 2.1 The convergentist lineage\n\n**Parfit.** The most ambitious modern convergence claim is Derek Parfit's in *On What Matters* (2011): that Kantians, contractualists, and rule consequentialists, when their theories are properly developed, \"are climbing the same mountain on different sides\" and converge on a single set of deontic principles — the \"Triple Theory\" ([Baumann 2021](https://doi.org/10.1007/s10677-021-10161-z); [Crisp 2020](https://doi.org/10.1007/s42048-020-00076-2)). Parfit's motivation is instructive: he held that persistent deep disagreement *under ideal conditions* would itself be evidence against objective normative truth, so convergence carried metaethical load ([Baumann 2021](https://doi.org/10.1007/s10677-021-10161-z)). The critical reception has been largely negative on the convergence claim specifically: Ridge finds the argument unsound ([Ridge 2009](https://doi.org/10.1111/j.1467-9329.2008.00418.x)); Walen argues the mountains are genuinely separate ([Walen 2024](https://doi.org/10.1163/17455243-20244110)); and a recurring objection is that agreement was purchased by modifying each theory until it no longer represents its tradition ([Baumann 2021](https://doi.org/10.1007/s10677-021-10161-z); [Joseph, \"Parfit's 'Triple Theory' and its Troubles\"](https://soar.suny.edu/server/api/core/bitstreams/c18f3b74-d1b2-46a5-81c3-c0f212b006ef/content)). *Confidence that Parfit's convergence proof is regarded as unsuccessful by most commentators: high.*\n\n**Michael Smith.** Smith's *The Moral Problem* (1994) makes convergence definitional: moral facts are facts about the desires that *all* fully rational agents would converge on, and moral objectivity stands or falls with that convergence ([Smith 1994](https://scispace.com/papers/the-moral-problem-1d49369ave)). Smith is explicit, in his own recent restatement, that this is an empirical hostage: \"if no such demonstration is to be had, then we will have to conclude that the concept... is not instantiated\" — i.e., error theory ([Smith, \"The Refined Moral Problem\"](http://www.filosofiskanotiser.com/Smith.pdf); [Smith 2024](https://doi.org/10.5937/bpa2437007s)). His positive argument — that moral argument's historical tendency to elicit agreement is best explained by convergence on unobvious a priori moral truths — is widely regarded as under-supported ([Sobel, PhilArchive](https://philarchive.org/archive/SOBDTD); [Enoch 2007](https://onlinelibrary.wiley.com/doi/10.1111/j.1468-0149.2007.00434.x); [Dalsotto 2018](https://doi.org/10.21680/1983-2109.2018v25n47id13254)). Smith's thesis is structurally identical to the platform's wager — reasoned criticism erodes idiosyncratic intrinsic desires until only shareable ones remain — and the philosophical debate has established only that this bet cannot be settled from the armchair. *Confidence: high.*\n\n**Railton and naturalist realism.** Peter Railton's reforming naturalism (\"Moral Realism,\" 1986) grounds moral facts in what an agent's fully-informed, fully-rational counterpart would want them to want, aggregated into \"social rationality\" ([Railton 1986](https://philpapers.org/rec/RAIMR)). Notably, Railton — an arch-realist — explicitly builds in relationality: \"by the nature of these criteria no one kind of life is likely to be appropriate for all individuals and no one set of norms appropriate for all societies\" ([Railton 1986, PDF](https://philpapers.org/archive/RAIMR.pdf)). Even the friendliest naturalist realism promises objective *criteria*, not uniform convergence on one way of life. Richard Boyd and David Brink in the same tradition argue that much apparent moral disagreement would resolve with agreement on non-moral facts — the canonical philosophical statement of C2 (surveyed in [SEP, \"Moral Disagreement\"](https://plato.stanford.edu/entries/disagreement-moral/)).\n\n**Habermas.** Discourse ethics makes convergence procedural: a norm is valid only if \"all affected can accept the consequences and side effects its general observance can be anticipated to have for the satisfaction of everyone's interests\" in a discourse free of coercion where \"nothing but the force of the better argument prevails\" ([SEP, \"Jürgen Habermas\"](https://plato.stanford.edu/entries/habermas/); [IEP, \"Habermas\"](https://iep.utm.edu/habermas/)). This is the closest philosophical ancestor of a deliberation platform: validity *just is* what ideal deliberation converges on. The standard critiques — the derivation of (U) has never been made fully rigorous ([SEP](https://plato.stanford.edu/entries/habermas/)), and consensus-orientation may paper over legitimate difference — mean discourse ethics *assumes* rather than *proves* that ideal discourse converges. *Confidence: high.*\n\n### 2.2 The pluralist counter-tradition\n\n**Rawls — the pivotal witness.** Rawls is often mis-slotted as a convergentist because of \"overlapping consensus.\" The record is the opposite: the mature Rawls holds that the \"burdens of judgment\" — the ordinary hazards of assessing evidence, weighing considerations, and interpreting concepts — \"all but preclude reasoned convergence on fundamental and comprehensive principles about how to live,\" making reasonable pluralism a *permanent* feature of free societies, erasable only by \"the oppressive use of state force\" ([IEP, \"Rawls\"](https://iep.utm.edu/rawls/)). The public-reason tradition generalizes the point: deep disagreement \"arises as a result of the normal functioning of human reasoning under reasonably favorable conditions\" ([SEP, \"Public Reason\"](https://plato.stanford.edu/entries/public-reason/); [SEP, \"Public Justification\"](https://plato.stanford.edu/entries/justification-public/)). Rawls's constructive move — converge on a *political* conception each person reaches from within their own comprehensive doctrine — is mid-level convergence without foundational convergence: arguably the most influential position in contemporary political philosophy, and a considered rejection of strong C3. *Confidence: high.*\n\n**Williams.** Bernard Williams's *Ethics and the Limits of Philosophy* (1985) supplies the sharpest conceptual challenge: in science \"we can coherently hope\" for \"a convergence on an answer where the best explanation of the convergence involves the idea that the answer represents how things are\"; in ethics, he argues, no such hope is coherent, because thin ethical concepts are not \"world-guided\" — if ethical convergence happened, its best explanation would be social and historical, not truth-tracking ([Cambridge, \"Science, ethics, and objectivity\"](https://www.cambridge.org/core/books/world-mind-and-ethics/science-ethics-and-objectivity/3E0AF26C418C574FEFCA248D9A386280); [Moore, \"Williams on Ethics, Knowledge, and Reflection\"](https://users.ox.ac.uk/~shug0255/pdf_files/williams-on-ethics-knowledge-and-reflection.pdf)). Note the scope: Williams does not deny convergence may *occur*; he denies it would *certify truth*. A platform can accept the point and still value convergence as agreement.\n\n**Berlin, Raz, Chang — the structure of the residue.** Isaiah Berlin's value pluralism holds that ultimate values are plural, often incompatible, and sometimes incommensurable: \"there is no objective procedural rule that enables us to balance one value against the other in such a conflict\" ([Isaiah Berlin Online](https://isaiah-berlin.wolfson.ox.ac.uk/isaiah-berlins-key-idea)). Joseph Raz and others developed incommensurability into a mainstream position in value theory ([SEP, \"Incommensurable Values\"](https://plato.stanford.edu/entries/value-incommensurable/)). Ruth Chang's refinement matters most for Deliberus: she argues most alternatives *are* comparable (against radical incommensurability), but that hard choices exhibit a fourth value relation, \"parity\" — neither better, nor worse, nor equal — and that such choices are \"not hard because of our ignorance\": more information does not resolve them ([Chang, \"The Possibility of Parity,\" Ethics 2002](https://philosophy.rutgers.edu/images/documents/publications/chang-POSSIBILITY_PARITY.pdf); [Chang, \"Hard Choices,\" 2017](https://doi.org/10.1017/apa.2017.7)). If parity is real, then *some* weighting disagreements survive full decomposition and full information as a matter of the structure of value, not as residual confusion. This is the most precise standing objection to C3. *Confidence that parity/incommensurability is a live, well-defended mainstream position: high. Confidence that it is true: genuinely contested.*\n\n**Wong — the sophisticated middle.** David Wong's \"pluralistic relativism\" (*Natural Moralities*, 2006) holds that universal constraints rooted in human nature and morality's function (enabling cooperation and individual flourishing) rule out most candidate moralities — but under-determine a unique winner: multiple true moralities remain, which \"often share some values (such as individual rights and social utility), but assign them different priorities\" ([SEP, \"Moral Relativism\"](https://plato.stanford.edu/entries/moral-relativism/); [Wong 2006](https://doi.org/10.1093/0195305396.001.0001)). Wong essentially predicts the empirical pattern documented in §4: universal structure, variable weightings.\n\n**Street — cutting both ways.** Sharon Street's Darwinian Dilemma argues that evolutionary forces \"have played a tremendous role in shaping the content of human evaluative attitudes,\" and that realists can neither deny a relation between those forces and evaluative truth (on pain of global moral skepticism) nor affirm a tracking relation (on pain of scientific implausibility) ([Street 2006](https://philpapers.org/rec/STRADD); [full text](https://harjitbhogal.com/Flukes2024/Street%202006%20-%20A%20Darwinian%20dilemma%20for%20realist%20theories%20of%20value.pdf)). For the wager this cuts three ways. *For* convergence-as-agreement: Street's premise is that selection gave humans deeply *shared* evaluative starting points (survival, kin care, reciprocity) — exactly what the cross-cultural data show (§4.4). *Against* convergence-as-truth: genealogically explained agreement does not certify mind-independent correctness (converging with Williams). *Against* guaranteed convergence: on Street's constructivism, full rationality corrects only incoherence relative to an agent's standpoint, so an ideally coherent agent with alien values remains possible (debate surveyed in [\"Evolutionary arguments against moral realism\"](https://pmc.ncbi.nlm.nih.gov/articles/PMC6245095/)). *Confidence in this reading: high.*\n\n---\n\n## 3. Idealized observers, full information, and formal results\n\n### 3.1 Ideal-observer and full-information accounts\n\nRoderick Firth defined moral rightness by the reactions of an \"ideal observer\" — omniscient, omnipercipient, disinterested, dispassionate, consistent, otherwise normal ([Firth 1952](https://www.jstor.org/stable/2103988)). Richard Brandt's *A Theory of the Good and the Right* (1979) operationalized idealization as \"cognitive psychotherapy\": rational desires are those that survive vivid, repeated exposure to all relevant information. Railton's ideal-advisor variant (§2.1) is the most developed successor.\n\n**The consensus verdict: these accounts do not deliver provable convergence, and the 1988–1995 critique wave is now the standard view.** The core problems, none fully answered:\n\n1. **Idealization changes the subject.** \"Fully informing people would change their motivations in ways that have nothing to do with the information itself\" ([Loeb 1995](https://gwern.net/doc/philosophy/ethics/1995-loeb.pdf)); the idealized self \"though purportedly you, may not be someone whose judgments you would recognize as authoritative\" ([Rosati 1995](https://gwern.net/doc/philosophy/ethics/1995-rosatigood.pdf)).\n2. **Full information may be impossible for perspectival beings.** Rosati argues that \"because of what it is like to be a person and to have a perspective, it appears that no person can be fully informed\" — vividly appreciating one possible life degrades one's ability to appreciate another ([Rosati 1995](https://gwern.net/doc/philosophy/ethics/1995-rosatigood.pdf); [Sobel 1994](https://philpapers.org/rec/SOBFIA-2)).\n3. **No uniqueness guarantee.** Nothing in the machinery guarantees that all ideal observers, or all idealized selves, land on the same verdicts; Firth had to stipulate agreement, and successors face a trilemma between circularity, redundancy, and loss of normative force ([Zhuang 2025](https://www.sav.sk/journals/uploads/03090810orgf.2025.32102.pdf); [Shemmer, Utilitas](https://www.cambridge.org/core/journals/utilitas/article/abs/full-information-wellbeing-and-reasonable-desires/EABC8626F6C027626BF6651B183E225E); Velleman's \"Brandt's Definition of 'Good'\" launched the wave — cited in [Sobel](https://philarchive.org/archive/SOBDTD)).\n\nRosati's conclusion is the right one for a deliberation platform: the fully-informed standpoint \"may still serve as a useful regulative ideal in theoretical inquiry,\" even though \"no account that attempts to embody it can be acceptable\" as a definition ([Rosati 1995](https://gwern.net/doc/philosophy/ethics/1995-rosatigood.pdf)). *Confidence: high.*\n\n### 3.2 Formal results\n\nThere is one genuinely strong formal convergence result, and it is about *beliefs*, not values. **Aumann's agreement theorem** (1976): Bayesian agents with common priors whose posteriors are common knowledge must have equal posteriors — \"people with the same priors cannot agree to disagree,\" even on different private information ([Aumann 1976](https://www.haverford.edu/sites/default/files/Aumann1976.pdf)). Successor results cover communication dynamics, approximate common knowledge, and computationally limited \"Bayesian wannabes,\" for whom persistent disagreement must trace to differing priors or computation, never information alone ([Hanson, \"For Bayesian Wannabes\"](https://mason.gmu.edu/~rhanson/disagree.pdf); [Hanson, \"Disagreement is unpredictable\"](https://mason.gmu.edu/~rhanson/unpredict.pdf)). Cowen and Hanson draw the sharp corollary: \"honest truth-seeking agents with common priors should not knowingly disagree,\" so typical persistent disagreement signals either non-common priors (usually self-favoring ones) or non-truth-seeking behavior ([Cowen & Hanson, \"Are Disagreements Honest?\"](https://hanson.gmu.edu/deceive.pdf)).\n\nThree caveats govern the transfer to the moral domain. (1) The theorem concerns propositions with truth values under shared priors; applying it to value questions requires cognitivism plus common evaluative priors — the very things in dispute. (2) The common-prior assumption is itself philosophically contested. (3) No analogous theorem exists for preferences or weightings; Chang's parity (§2.2) is a direct argument that the value domain lacks the total ordering such a theorem would need. **Net: under strong idealization, factual and instrumental disagreement (C2) is provably unstable; nothing comparable exists for terminal weightings (C3).** *Confidence: high.*\n\n### 3.3 The empirical analogue: adversarial collaboration\n\nAdversarial collaboration — disagreeing experts co-designing tests — is the closest real-world approximation of idealized joint inquiry. The track record: Kahneman reported his collaborations ended with \"new facts accepted by all, narrowed differences of opinion, and considerable mutual respect\" ([APA Monitor 2025](https://www.apa.org/monitor/2025/04-05/adversarial-research-collaboration)), while warning that \"even successful collaborations will end with few minds changing\" ([Kahneman, Edge lecture](https://www.edge.org/adversarial-collaboration-daniel-kahneman)). The Kahneman–Killingsworth happiness dispute was fully *resolved* this way in 2023 — the crux proved to be a data artifact ([Globe and Mail account](https://www.theglobeandmail.com/canada/article-scientists-pursuit-of-happiness-debate/)); advocates argue the method is needed precisely because current norms let \"contradictory conclusions... persist for decades with little to no convergence\" ([Clark, Costello, Mitchell & Tetlock 2022](https://psycnet.apa.org/doiLanding?doi=10.1037/mac0000004&)). The honest summary: structured adversarial inquiry reliably *narrows* disagreement, dissolves it most often when the crux was factual, and rarely produces capitulation on interpretive or evaluative cruxes. *Confidence: high.*\n\n---\n\n## 4. Moral psychology: are weightings traits, or artifacts of framing and information?\n\n### 4.1 Moral foundations: much less trait-like than advertised\n\nMoral Foundations Theory (MFT) popularized the picture of liberals and conservatives running on stably different moral \"taste receptors\" ([Graham, Haidt & Nosek 2009](https://doi.org/10.1037/a0015141)). Three findings complicate the trait reading:\n\n- **Panel data invert the causal arrow.** Longitudinal studies find moral foundations *less* temporally stable than political attitudes, with little evidence of genetic influence, and ideology predicting later foundations better than the reverse — \"Ideology Justifies Morality,\" i.e., foundations are not behaving like \"stable, dispositional traits\" ([Hatemi, Crabtree & Smith 2019](https://doi.org/10.1111/ajps.12448); [authors' summary](https://ajps.org/2019/07/31/ideology-justifies-morality/); building on [Smith et al. 2017](https://doi.org/10.1111/ajps.12255)).\n- **Framing moves foundation scores.** Survey experiments show MFQ responses shift with political context cues ([Ciuk 2018](https://doi.org/10.1177/2053168018781748)).\n- **The meta-analytic picture is real but heterogeneous.** Across 89 samples the liberal–conservative foundation differences replicate, but effect sizes from the flagship self-selected dataset appear inflated, and associations vary across countries and subcultures ([Kivikangas et al. 2021](https://doi.org/10.1037/bul0000308)). MFT's own authors have since rebuilt the instrument (MFQ-2), finding that \"the nomological network of moral foundations varied across cultural contexts\" ([Atari et al. 2023](https://doi.org/10.1037/pspp0000470)).\n\n**Reading for the wager:** foundation *weightings* are partly downstream of ideology, identity, and framing — i.e., of things that argument, evidence, and reframing can reach — rather than fixed pre-political bedrock. This is more convergence-friendly than the popular reading of MFT. It does not show weightings are *fully* malleable. *Confidence: moderate-to-high (the panel studies are strong but contested by MFT proponents).*\n\n### 4.2 Does reflection actually shift moral judgment?\n\n- **Reflection manipulations work.** Inducing reflectiveness (via the Cognitive Reflection Test) increased utilitarian responding; and in the famous consensual-incest vignette, a *strong* argument was more persuasive than a weak one — but only when a delay forced deliberation ([Paxton, Ungar & Greene 2012](https://doi.org/10.1111/j.1551-6709.2011.01210.x)). Argument quality has causal force when reflection is scaffolded — a direct empirical warrant for deliberation infrastructure.\n- **But interpret the direction cautiously.** \"Utilitarian\" dilemma responses do not indicate impartial benevolence — they correlate with subclinical psychopathy and rational egoism, not with real-world concern for the greater good ([Kahane et al. 2015](https://doi.org/10.1016/j.cognition.2014.10.005); [Kahane et al. 2018](https://kar.kent.ac.uk/73822/)). Reflection changes judgments; it does not follow that it moves them toward *moral truth*.\n- **Moral dumbfounding is far weaker than its fame.** The original demonstration was an unpublished N=30 study. Royzman, Kim & Leeman found most condemners actually maintained harm beliefs (reasonably, on their evidence), and with beliefs and normative standards properly controlled, \"a more rigorous assessment procedure yielded a dumbfounding estimate of about 0\" ([Royzman et al. 2015](https://www.sas.upenn.edu/~baron/journal/15/15405/jdm15405.html)). Direct replications find dumbfounded responding exists but at rates highly sensitive to measurement ([McHugh et al.](https://www.cillianmchugh.com/publications/pubs/searching-for-moral-dumbfounding/)). The rationalist counter-tradition — Mercier and Sperber's argumentative theory (reasoning evolved for social argument evaluation and performs well in dialogic contexts), plus related arguments popularized by Bloom and Pinker — has on balance strengthened against the intuitionist \"reason is a press secretary\" picture since 2015 ([Mercier & Sperber 2011](https://www.cambridge.org/core/journals/behavioral-and-brain-sciences/article/abs/why-do-humans-reason-arguments-for-an-argumentative-theory/53E3F3180014E80E8BE9FB7A2DD44049)). *Confidence: high that dumbfounding was overclaimed; moderate on the strength of the rationalist rebound.*\n- **Values respond durably to confronted inconsistency.** The oldest and most striking evidence: Rokeach's value self-confrontation experiments made participants aware of inconsistencies within their own value-attitude systems (e.g., ranking freedom far above equality); this produced value re-ranking and *behavioral* change (NAACP joining, curriculum enrollment) persisting 15–21 months ([Rokeach 1971](https://doi.org/10.1037/h0031450); replications and boundary conditions reviewed in [Grube, Mayton & Ball-Rokeach 1994](https://spssi.onlinelibrary.wiley.com/doi/10.1111/j.1540-4560.1994.tb01202.x)). Decomposition-plus-confrontation is a value-change mechanism with five decades of evidence. *Confidence: moderate (old paradigm, but replicated within its literature).*\n- **Reframing across value languages persuades.** Arguments reframed to fit the *audience's* moral values (e.g., environmental arguments in purity terms, or anti-candidate arguments in loyalty terms) outperform arguments in the speaker's own values, across many polarized topics ([Feinberg & Willer 2015](https://doi.org/10.1177/0146167215607842); [Voelkel & Feinberg 2017](https://doi.org/10.1177/1948550617729408); review: [Feinberg & Willer 2019](https://doi.org/10.1111/spc3.12501)). Translation across worldview lenses is an empirically validated bridging mechanism.\n\n### 4.3 Facts and values are entangled — in both directions\n\nThe wager's C2 assumes disagreement often sits on factual/instrumental beliefs. The philosophical statement of this (Boyd, Brink) is standard ([SEP, \"Moral Disagreement\"](https://plato.stanford.edu/entries/disagreement-moral/)). But the empirical literature adds a crucial complication: **the causal arrow also runs backwards.** People align their factual beliefs about consequences with their prior moral evaluation of an act's inherent wrongness; reading essays about the inherent (im)morality of capital punishment changed beliefs about its *costs and benefits* even though no consequence information was supplied ([Liu & Ditto 2013](https://journals.sagepub.com/doi/10.1177/1948550612456045)). Merely making values salient can causally shift factual beliefs, and partisan factual disagreement arises from rationalization of common information ([Bruckmeier et al., motivated political reasoning](https://rationality-and-competition.de/wp-content/uploads/discussion_paper/510.pdf); [Lauderdale](https://www.cambridge.org/core/journals/political-science-research-and-methods/article/abs/partisan-disagreements-arising-from-rationalization-of-common-information/25DBC5ABC92E00C994E8BA06A25EA512)). So \"fix the facts and values-agreement follows\" is too simple: fact-fixing and value-clarification must iterate, because motivated cognition regenerates congenial \"facts.\" *Confidence: high.*\n\n### 4.4 Universal structure, variable weights: the strongest empirical pattern\n\nAcross independent research programs, the same shape recurs:\n\n- **Schwartz values.** The circular motivational structure of basic human values replicates across cultures ([Sagiv & Schwartz 2022, Annual Review of Psychology](https://doi.org/10.1146/annurev-psych-020821-125100)), and — the underappreciated result — average value *hierarchies* are strikingly similar worldwide: benevolence, self-direction, and universalism at the top; power, tradition, stimulation at the bottom; \"value hierarchies of 83% of samples correlate at least .80 with this pan-cultural hierarchy\" ([Schwartz & Bardi 2001](https://doi.org/10.1177/0022022101032003002)).\n- **Morality-as-cooperation.** In the ethnographic records of 60 societies, seven cooperative behaviors (help kin, help group, reciprocate, be brave, defer to superiors, divide fairly, respect possession) have *uniformly positive* moral valence — zero societies treat them as bad ([Curry, Mullins & Whitehouse 2019](https://doi.org/10.1086/701478); [Oxford summary](https://www.ox.ac.uk/news/2019-02-11-seven-moral-rules-found-all-around-world)).\n- **Sacrificial dilemmas at scale.** Across 70,000 participants in 42 countries, trolley-type judgments show \"a universal qualitative pattern of preferences together with substantial country-level variations in the strength of these preferences\" ([Awad et al. 2020, PNAS](https://www.pnas.org/doi/10.1073/pnas.1911517117)); the 40-million-decision Moral Machine dataset likewise found globally shared directional preferences with three cultural clusters differing in *degree* ([Awad et al. 2018, Nature](https://pubmed.ncbi.nlm.nih.gov/30356211/)).\n\n**This is the empirical core of the review: humans broadly share the *inventory* and even the rough *ordering* of values; cultures and individuals differ mainly in weightings, applications, and scope — and those differences are real but bounded.** The historical proof-of-concept for mid-level convergence without foundational convergence is the Universal Declaration of Human Rights: Maritain famously reported that the drafters could agree on the rights \"provided no one asks us why\" — practical convergence atop irreconcilable justifications ([UNESCO 1948 symposium, Maritain's introduction](https://e-docs.eplo.int/phocadownloadpap/userupload/aportinou-eplo.int/Human%20rights%20comments%20and%20interpretations.compressed.pdf); [drafting history](https://doi.org/10.14288/1.0090389)). *Confidence: high.*\n\n### 4.5 Deliberation converges — under the right conditions\n\nWhether discussion produces convergence or polarization is strongly condition-dependent. Like-minded, unstructured groups predictably move to \"a more extreme point in the direction indicated by their own predeliberation judgments\" — the law of group polarization ([Sunstein 1999/2002](https://chicagounbound.uchicago.edu/cgi/viewcontent.cgi?article=1541&context=law_and_economics)). But structured deliberation with balanced materials, diverse participants, and moderation shows the opposite: in the \"America in One Room\" national field experiment (500+ voters, control group), deliberators showed \"large, depolarizing changes in their policy attitudes and large decreases in affective polarization,\" with two-sided depolarization on all 26 initially polarized proposals among those starting at the extremes ([Fishkin, Siu, Diamond & Bradburn 2021, APSR](https://doi.org/10.1017/s0003055421000642)). The argumentative theory explains both regimes with one mechanism: individual reasoning is confirmation-biased, but evaluating others' arguments in genuinely diverse exchange is where reasoning performs well — groups outperform individuals when opinions are diverse, and polarize when they are not ([Mercier & Landemore 2012](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1707029); [Mercier & Sperber 2011](https://www.dan.sperber.fr/wp-content/uploads/2011_mercier_why-do-humans-reason.pdf)). *Confidence: high.*\n\n### 4.6 Semantic confusion is a documented, tractable component\n\nThe philosophical toolkit here is Chalmers's \"method of elimination\" for verbal disputes: bar the contested term and check whether a substantive disagreement survives in more basic vocabulary — noting that \"many philosophical disagreements are at least partly verbal\" ([Chalmers 2011, Philosophical Review](https://philpapers.org/rec/CHAVD); [full text](https://consc.net/papers/verbal.pdf)). This is in effect a manual specification of a clarification pipeline, with Chalmers's own caution built in: elimination usually reveals a *residual* substantive dispute in more basic (\"bedrock\") terms. Semantic clarification shrinks, localizes, and purifies disagreements — it rarely eliminates them. *Confidence: high.*\n\n---\n\n## 5. Where the academic weight actually lands\n\nAn honest distribution, not just the strongest arguments:\n\n- **Metaphysics: convergence-friendly.** 62% of professional philosophers accept or lean toward moral realism, 26% anti-realism ([2020 PhilPapers Survey](https://survey2020.philpeople.org/survey/results/all); [Bourget & Chalmers 2023](https://journals.publishing.umich.edu/phimp/article/id/2109/)). Most experts think there is something to converge *on*.\n- **First-order theory: stubbornly split.** The same survey shows deontology (32%), consequentialism (31%), and virtue ethics (37%) in a rough three-way tie (accept-or-lean, overlapping answers permitted) — essentially unchanged since 2009. Parfit's attempted proof that the traditions secretly agree is admired and mostly judged unsuccessful (§2.1).\n- **Idealization: judged insufficient for uniqueness.** The Sobel–Rosati–Loeb–Velleman critique of full-information accounts is the textbook position (§3.1); ideal-observer convergence is stipulated, not derived.\n- **Political philosophy: reasonable pluralism is the mainstream.** Rawls's burdens of judgment — reasonable people persistently disagree on comprehensive doctrines *under favorable conditions* — is probably the single most widely endorsed position surveyed here (§2.2).\n- **Moral psychology: shared structure, malleable-but-not-fully-malleable weights.** Universal value inventories and near-universal aggregate orderings (§4.4); weightings that respond to reflection, confrontation, reframing, and ideology (§4.1–4.2); motivated cognition that entangles facts with values in both directions (§4.3); deliberation that converges under structured-diverse conditions and diverges otherwise (§4.5).\n\n**Overall verdict.** The proposition \"good-hearted people converge under idealized reasoning\" commands neither consensus nor refutation. The *strong* version — full convergence, including terminal weightings, given enough decomposition and information — is a minority position among experts; distinguished defenders exist (Parfit, Smith, Habermas, Boyd), but the modal expert expects a residue of reasonable pluralism (Rawls, Wong, Chang, Berlin), and the best formal results cover beliefs, not values. The *moderate* version — most disagreement is semantic, factual, or instrumental; decomposition plus structured diverse deliberation substantially narrows disagreement; convergence is strong at the level of basic values and mid-level principles and weakest at final weightings — is well supported by both philosophy and data. The wager, stated as a bet about *where the residue lies and how large it is*, is a respectable live hypothesis that no one has earned the right to assert as established — which is precisely what makes it worth instrumenting. *Confidence in this meta-assessment: high.*\n\n---\n\n## 6. Implications for Deliberus\n\n1. **The platform's mechanisms target exactly the layers where convergence is best-evidenced.** Contested-concept detection and interpretation checkpoints operationalize Chalmers's method of elimination (C1); evidence-linking and critical questions target the factual/instrumental layer where Aumann-type pressure and adversarial-collaboration experience show disagreement is genuinely unstable (C2); decomposition-with-confrontation is the Rokeach mechanism, the only intervention with long-horizon evidence of durable value change (C3, weak form). The architecture is not merely *compatible* with the literature — it is what the convergence-friendly literature repeatedly recommends and rarely builds.\n2. **Expect and instrument a convergence gradient.** Predicted convergence yield, highest to lowest: verbal disputes → facts → instrumental/feasibility beliefs → mid-level principles (UDHR-style overlapping consensus) → terminal weightings (parity residue). Measuring convergence claim-by-claim along this gradient, rather than asserting it globally, turns a philosophical stalemate into an empirical research program the graph is uniquely positioned to run: the wager becomes falsifiable at scale.\n3. **Legibility is the guaranteed floor; convergence is the measurable upside.** Chang's parity and Rawls's burdens of judgment imply some fully-decomposed, fully-informed disagreements will remain — and will be *visible as such*. A decomposed disagreement whose residue is two clean weighting-atoms with acknowledged parity is a success state, not a failure state. Design the UX so that \"we now agree on 47 of 50 claims and can name the two weightings we differ on\" reads as the achievement it is.\n4. **Structure is load-bearing.** The same human material converges under Fishkin-style conditions (balanced materials, diversity, moderation) and polarizes under Sunstein-style conditions (homogeneous, unstructured). A deliberation platform is not a neutral pipe: its design determines which regime users are in, and interlocutor diversity around a claim neighborhood is a first-class variable worth monitoring.\n5. **Worldview-lens translation is empirically validated.** Moral reframing (Feinberg & Willer) shows arguments land when expressed in the audience's value language. The lens/bridging features have direct experimental support, and the bridging signal (arguments compelling across clusters) is the platform's most distinctive measurable quantity.\n6. **Iterate facts and values; never assume one-pass resolution.** Because moral evaluation reshapes factual belief (Liu & Ditto), a single fact-check pass will under-deliver; the graph should expect values-driven regeneration of congenial factual claims and support repeated re-confrontation.\n7. **C4 — \"better-informed and better-reasoned\" — is defensible without settling metaethics.** Even committed anti-realists (Street) grant standpoint-internal correction: consistency, evidential support, answered critical questions, and surviving decomposition are quality dimensions every metaethical camp accepts, and QBAF-style gradual semantics operationalizes them. The platform can assert \"this position is better-reasoned\" with full confidence while staying agnostic about whether better-reasoned positions must ultimately agree.\n8. **Persistent disagreement is diagnostic signal.** The Aumann/Hanson results imply that, among honest truth-seekers, disagreement surviving full information-sharing localizes in priors, computation, or values. A decomposition graph makes that localization explicit — turning \"we disagree\" into \"we disagree *here*, and it is a prior/weighting/fact of this kind\" — valuable output even when convergence fails.\n\n---\n\n## 7. Strongest objection to my own conclusions\n\nThe synthesis above leans optimistic in three ways that a skeptic should press. First, **selection effects**: the encouraging evidence (Fishkin's deliberators, Rokeach's students, adversarial collaborators, lab reflection effects) comes from structured settings with cooperative participants — but the disagreements a deliberation platform exists for are precisely those *selected for resistance*: identity-laden, motivated, adversarial. Evidence that deliberation works where it works only weakly predicts it works where it is most needed. Second, **the aggregation fallacy risk**: the pan-cultural value hierarchy and the seven universal moral rules are aggregate-level findings; individuals argue as individuals, and much of the variation in value priorities lies within societies rather than between them — so \"humanity shares a value ordering\" may say little about whether *these two disputants* do. Third, and most uncomfortably: **philosophers are the best decomposers in history, argue under unusually favorable conditions, and have not converged** — the PhilPapers three-way normative-ethics split has barely moved in a decade, and the field's most celebrated convergence proof (Parfit's) is widely judged to fail. If decomposition plus idealized reasoning sufficed for convergence, professional philosophy is where it should have shown up first. The convergentist's best reply — that theoretical traditions increasingly agree in *verdicts* even where foundations differ, and that ordinary practical disputes are less foundational than philosophy's — is plausible but unproven. The honest posture is the one this document recommends: treat convergence as a measurable hypothesis with a predicted gradient, and let the graph report where it actually happens.\n"}