{"path":"research/good-hearts-and-permissivism.md","content":"# Good hearts and permissivism: two holes in the convergence line, and the answers he would give\n\n**Date**: 2026-10-25, evening. **Status**: a session's reading, unruled. Nothing here changes a ruled line; what\nsurvives comes to him as candidate lines of their own (§ 5), as he asked.\n\n## 1. His questions\n\nThe convergence conviction was redrafted in his own words the same evening and ruled. Two of its phrases drew\nquestions while it was being put to him:\n\n- On *\"hinge/parity/permissivism ceilings count against it\"*: *\"One idea: I'm not sure this holds up as counting\n  against? Poke holes in this for me please.\"*\n- On *\"they have good hearts\"*: *\"Poke holes in this as well, perhaps based on Jonathan Haidt's research and such, idk.\n  Then try to see how you can rebut all critique as you predict I would, based on what Deliberus could become.\"*\n- His ruling on the redraft, with the same thought: *\"Ruled and blessed, we can put any rebuttals etc as their own lines\n  perhaps?\"*\n\nRead-status of the outside sources: Waytz, Young and Ginges, Sanz-García et al. and the Skitka review were read at\nabstract level on 2026-10-25; Haidt, Choi and Bowles, Darley and Batson, Blackwell and Dubins, White and Schoenfield\nare standard citations recalled, not read for this note. The corpus notes cited were read in their sections.\n\n## 2. Permissivism\n\n**What it is.** Rational permissivism is the thesis that one body of evidence can rationally permit more than one\nattitude towards a claim (White 2005 argues against, Schoenfield 2014 for). Its rival, uniqueness, says the evidence\nfixes one rational attitude. His own founding position leans to uniqueness: ledger P2, *\"if there's inductive evidence\nfor any claim, that's what determines the probability assessment … So that was in my mind not different for different\npeople ideally. But of course maybe it is irreducibly subjective\"*.\n\n**Where the rule came from.** The July stress test (`philosophical-foundations-stress-test.md` § 4) split the\npermissive zone into parity, indeterminacy, incomparability and permissivism, and proposed that parity and permissivism\ncount as ceilings on convergence. The red-team synthesis asked him to ratify that (item 2), and `convergence.md`\n§ The accounting states it. **But the same file says, further down, that which of these counts against the wager is an\nopen ruling (ledger P2)**, so the canon line has stated as settled what the corpus's own home for it calls open. The\nreason for the rule was sound: a wager whose accounting scores every ceiling as work remaining cannot lose. The holes\nbelow are in what it counts, not in the need to count.\n\n**Hole 1: starting beliefs wash out where evidence keeps arriving.** The merging-of-opinions theorem (Blackwell and\nDubins 1962) says that people who start from different priors, neither ruling out what the other allows, come to agree\nas shared evidence accumulates. So a permissive start counts against convergence only where evidence stops arriving,\nor where one prior excludes what the other allows. **Provenance** (his question, *\"What's the provenance of this idea?\"*; recalled, not read for this note): de Finetti's work on exchangeability (1937) and Savage's *Foundations of Statistics* (1954) argued that people with different personal probabilities come to agree as shared observations accumulate; Doob (1949) proved posteriors consistent; Blackwell and Dubins (1962, *Annals of Mathematical Statistics*) proved the merging theorem in general form, and Gaifman and Snir (1982) carried it to rich languages. Its philosophical ancestor is Peirce's truth as the limit of inquiry, which `convergence.md` names as the wager's root. The corpus's home is `starting-values-and-the-problem-of-the-prior.md` § 2.4, with the caveat that these are limit results with no direct force for finite data. Two conditions carry the theorem: priors that rule out nothing the other allows, and **agreement on what the evidence means** (the same likelihoods). The second is where people actually part, which is why the agreement Deliberus seeks on what the evidence says (ledger Q15, open) matters more than the theorem. The corpus already says the narrower thing:\n`starting-values-and-the-problem-of-the-prior.md` § 6.6, *a region that stays prior-dependent however much evidence\narrives is a permissive horizon, counted against the wager*.\n\n**Hole 2: Deliberus's own design takes the person out of the prior.** By BC1 (ruled: a claim's prior is read from what\nit says, through counted reference classes, for anyone reading it), and by the agreement that is reachable on what the\nevidence says (ledger Q15), the graph computes the same posterior for every reader. Permissivism can enter only at\nthree places, each of them a named and arguable thing: the choice of reference class (a claim), the decision threshold\n(set by error costs and stored on no claim: `burden-of-proof-and-priors.md`), and the weights on values. So in\nDeliberus the numbers are not permissive by construction; what people do with them can differ.\n\n**⚠ Hole 3 withdrawn 2026-10-25 on his correction**, *\"That's not really the point of the convergence thesis/wager and I hope you see that?\"*: the wager is that reasons, decomposed deeply enough, converge on what is shared at depth; it never claimed that degrees of belief match, so the hole attacked a reading nobody holds. Kept below as written. **Hole 3 (withdrawn): counted against, it makes the wager lose by default.** Interpersonal permissivism about exact degrees of\nbelief is true nearly everywhere, since no two people's credences match to the decimal. If every permissive leaf\ncounts against convergence, the wager loses before anyone descends: the mirror image of the unfalsifiability the\naccounting rule was written to stop. A fair count needs a threshold, divergence that would change a verdict band or a\ndecision. Deliberus already decides whether a question is worth asking by whether its answer would cross a display\nband (`worth_asking`), which is the same threshold.\n\n**Hole 4: a band both sides grant is agreement on the map.** When both sides grant that the evidence permits either\nanswer, they agree on what the landscape is, in his words *\"share more landscape that's mappable\"*, and differ in the\npick. That is convergence of understanding with a difference of choice; scored as divergence it hides the agreement.\n\n**What survives at full strength.**\n\n- The strong headline. If rational people may differ permanently in standards at the bottom, the *same bottom* half of\n  the wager fails, as `convictions-across-scales.md` already found from outside: convergence onto a small basis, never\n  the same basis.\n- A permission must never be declarable at will. The corpus reads a permissive zone from what survives attack (ledger\n  O4, O11; `parity-and-the-zone-interior.md` § Addendum 2026-09-14), or *it's permitted* becomes a way to end an\n  argument without winning it.\n- The forecasting collaboration whose experts barely moved after 80 hours (`empirical-deliberation-convergence-evidence.md`,\n  the FRI residue) is a measured case of something keeping informed people apart. Hole 1 places it: forecasts of the\n  future are where evidence does not arrive in time.\n\n**The same split probably applies to parity** (both sides grant the options are on a par, and the commitment differs),\nwhich this note does not examine.\n\n## 3. \"They have good hearts\"\n\n**His sentence, at its home.** `vision.md` § The Founding Wager: *\"I believe the vast majority of people have good\nhearts. They interpret what's feasible and wise differently — through different experiences, different information,\ndifferent meanings hiding in the same words — but the care underneath is broadly shared.\"* The ruled canon line quotes\nhis message of 2026-07-04, put as a question: *\"For the vast majority of the population, they have good hearts. They\njust interpret what's feasible and wise differently?\"*\n\n**The holes.**\n\n1. **People differ in which goods count, not only in what is wise.** Moral foundations theory (Graham, Haidt and Nosek\n   2009) finds people weight care and fairness against loyalty, authority and sanctity differently, so a disagreement can\n   be about what counts as moral at all. Haidt's summary in *The Righteous Mind* (2012): morality binds and blinds; the\n   same commitments that bind a group blind it to the other.\n2. **Goodness is often parochial.** Altruism towards one's own and hostility to outsiders appear to have evolved\n   together (Choi and Bowles 2007). A heart can be good to us.\n3. **Good-hearted people reason towards their group's answer.** Identity-protective cognition (Kahan) shows that more\n   reasoning skill brings more polarization on identity-laden questions, so the different interpretations are not\n   innocent.\n4. **Moralized attitudes resist settlement.** Attitudes held as moral convictions resist procedural solutions and are\n   relatively immune to authority and peers (Skitka, Hanson, Morgan and Wisneski 2021, *Annual Review of Psychology*;\n   Napier and Tyler 2008 dispute the strong form).\n5. **Situations predict behaviour better than hearts.** Seminarians in a hurry passed a person in need on their way to\n   a talk about the Good Samaritan (Darley and Batson 1973).\n6. **The minority is small and concentrated.** Psychopathy is estimated at 4.5% of adults pooled across instruments,\n   1.2% on the strictest instrument, and higher in samples from organizations than from communities (Sanz-García et al.\n   2021). Small, and found where decisions are made.\n7. **Good hearts can want the same field.** A conflict of interest is not a failure of heart; the opacity conviction\n   says as much: understanding dissolves confusion-conflict and only clarifies interest-conflict.\n8. **\"Good\" by whose lights?** A heart good by care can be bad by loyalty, so the claim needs a measure that is not one\n   side's foundations.\n\n**Evidence for the claim.**\n\n- **Each side mistakes the other's heart.** In five studies of 661 American partisans, 995 Israelis and 1,266\n  Palestinians, each side credited its own group's aggression to love for its own and the other side's to hate; paying\n  for accuracy reduced the error (Waytz, Young and Ginges 2014, *PNAS*). The belief that the other side's hearts are bad\n  is measurably the commonest mistake.\n- Partisans imagine about twice as many extremists on the other side as there are (More in Common 2019;\n  `trust-economics-and-false-polarization.md`).\n- Moral foundations are less stable than political attitudes and partly follow ideology and framing (Hatemi, Crabtree\n  and Smith 2019; `moral-convergence-metaethics-and-psychology.md` § 4.1), so the weightings in hole 1 are partly\n  within reach of argument.\n- Seven cooperative behaviours are valued positively across 60 societies (Curry, Mullins and Whitehouse 2019).\n\n## 4. The answers he would give, from what Deliberus could become\n\nA session's prediction of his rebuttals, each with how far it holds.\n\n| Hole | His predicted answer | Holds? |\n|---|---|---|\n| 1, which goods count | Weighting is part of judging what is wise. Deliberus opens *\"we just weight things differently\"* into beliefs about consequences, postures towards risk, whose interests count and rules for adding up (`vision.md`), each arguable; foundations partly follow ideology and framing; moral reframing translates an argument across foundations (Feinberg and Willer). | Partly. A residue of weighting stays, as the forecasting collaboration showed. |\n| 2, parochial goodness | Who counts is itself a claim in the graph and can be argued; the motive-attribution studies show each side underrates the other's love for its own. | For the misperceived share. A real preference for one's own stays, as a stated position. |\n| 3, motivated reasoning | Curiosity reverses identity-protective reasoning (Kahan); reasons carried by a person's story move people where a group's flag does not (Kalla and Broockman). | If curiosity is recruited; the design has to do that. |\n| 4, moral conviction | Deliberus asks nobody to compromise; it maps. On sacred values, recognition rather than argument brings flexibility (Ginges and Atran), which is the attunement pole's work. | For legibility. It does not win convergence. |\n| 5, situations | Then the situation is the lever, and Deliberus is a situation: remove the constraint rather than repair the person, the harness conviction applied to people. | Strongly. The critique becomes support. |\n| 6, the minority | *\"The vast majority\"* carries the scope; anti-gaming is core work, and removing a payoff needs nobody's good heart. | Yes, with Porter's capture rider: the minority sits where agendas are set. |\n| 7, interests | The opacity conviction already concedes it; mapping still finds the moves both can gain from that misperception hid. | As a statement of scope. |\n| 8, whose lights | The claim is care for others' welfare in some form, which every foundation contains; and it is testable: if descents bottom out in indifference or malice rather than different beliefs, weights and circles of concern, the wager loses. | If a reading for *no shared care found* is built and counted. Not built. |\n\n## 5. What survives, as candidate lines (unruled, for him one at a time)\n\n1. **Good hearts, narrowed.** *Most disagreements, decomposed, show shared care held with different beliefs, weights and\n   circles of concern; the share that bottoms out in indifference or malice is small, and Deliberus counts it.* The word\n   that does not survive is *just*: a difference in weights or in who counts is more than interpretation, though each is\n   arguable.\n2. **The commonest error between sides is believing the other's heart is bad** (Waytz, Young and Ginges), and legible\n   reasons are its correction. A design target, demonstrated outside the project.\n3. **A reading the wager owes**: a descent that finds no shared care counts against convergence. Not built.\n4. **Permissivism, narrowed.** *Permissivism counts against convergence only where a claim stays prior-dependent however\n   much evidence arrives; a band both sides grant is agreement on the map and a difference in the pick, counted as\n   such.* If ruled, the convergence line's *permissivism* is narrowed to match, and `convergence.md` § The accounting\n   with it. Put to him the same evening in a plainer form (his reply, *\"This one first, please.\"*, read as this line\n   before the good-hearts line): *Permissivism, the thesis that one body of evidence can rationally permit more than\n   one belief, counts against convergence only where a claim's posterior still turns on a contested choice (of\n   reference class, decision threshold or value weight) however much evidence arrives; where both sides grant that\n   the evidence permits either answer, they agree on the map and differ only in the pick, and the horizon map scores\n   it that way.* Grade: predicted.\n\n## 6. The shared numbers and the lens are one structure (his prompt, 2026-10-25 evening)\n\nHis words, on the sentence that BC1 makes the graph's numbers the same for every reader and leaves divergence only\nin contested choices: *\"This relates closely to the attunement/worldview-lenses/what-if-simulating/mind-meld material.\nThink deeply about this.\"* A session's reading, unruled.\n\n**The connection.** `what-if-and-lenses.md` § 4 already gives a lens its anatomy: the places two minds part (AN11: a\nfact, a prior, a sense, a weighting, a standard, a stake), of which the first four can be *supposed* and recomputed by\nthe what-if view, and the last two can only be shown in the holder's words. BD25 says divergence that evidence cannot\nsettle lives in contested choices. These are the same list seen from two sides. **The graph's numbers are the commons,\nthe same for every reader; a lens is one person's declared setting of the contested choices; the what-if view\nrecomputes the commons under that setting; and the mind-meld shows two people the few choices on which their picks\npart.** Permissivism, in BD25's sense, is then exactly what the lens leaves standing after it has located everything:\nthe choices that arguing does not narrow.\n\n**What BC1 changes in the lens.** The lens note's prior dial reads *\"the base-rate claims the holder accepts\"* (Q35,\nopen). Since BC1 (ruled 2026-10-24: a claim's prior is read from what it says, through counted reference classes, for\nanyone reading it, and no model states a rate) a holder cannot hold a private rate; what a lens can suppose is **which\nreference class applies**, and that choice is itself a claim in the shared graph, arguable like any other. So the\nprior dial survives as a dial over reference classes, never over numbers.\n\n**What the session's list missed.** The supposable places include a word's **sense** (AM7, ruled: every use is an\nattributed vote on the meaning) and **what a holder finds in need of explaining** (AT7, open), and Q15 adds a\n**model**. BD25's parenthesis names three (reference class, decision threshold, value weight); read as examples it\nholds; read as a list it is short by three. The decision threshold is not in AN11's six at all: it is set by the costs\nof each error (`burden-of-proof-and-priors.md`), a weighting of outcomes rather than of reasons, and it too can be\nsupposed as claims about costs.\n\n**Where the convergence wager actually lives, then.** Not in degrees of belief matching (hole 3, withdrawn), but in\nwhether arguing the choices narrows them. Each choice is a claim in the shared graph: a reference class can be shown to\nfit or not, a sense can be shown to be a motte, a cost can be argued. The wager is that the set of choices on which\nholders still part, after the arguing, is small and named. The lens is the instrument that makes that set visible;\nthe horizon map is where its size is counted.\n\n**And attunement.** AN11's own line: *\"A mind-meld that shows only the first four shows two arguments. With the last\ntwo it shows two people.\"* The commons computes reasons; the lens shows causes beside them, in each holder's words, so\nthe permissive residue is met as two people's standing, not as an error to correct.\n\n**Candidate line (unruled):** *A worldview lens is one person's declared setting of the contested choices (reference\nclass, sense, weighting, what needs explaining, the costs of each error); the graph's numbers are the same for every\nreader, the what-if view recomputes them under a lens, and two lenses side by side show the few choices on which two\npeople part, each of them a claim that can be argued.*\n\n## Sources\n\n- Waytz, Young and Ginges 2014, *Motive attribution asymmetry for love vs. hate drives intractable conflict*, PNAS:\n  https://moralitylab.bc.edu/wp-content/uploads/sites/192/2011/10/PNAS-2014-Waytz.pdf\n- Sanz-García et al. 2021, *Prevalence of Psychopathy in the General Adult Population*, Frontiers in Psychology:\n  https://pmc.ncbi.nlm.nih.gov/articles/PMC8374040/\n- Skitka, Hanson, Morgan and Wisneski 2021, *The Psychology of Moral Conviction*, Annual Review of Psychology;\n  Napier and Tyler 2008 and Skitka and Mullen 2008, Social Justice Research.\n- Graham, Haidt and Nosek 2009; Haidt 2012, *The Righteous Mind*; Choi and Bowles 2007, Science; Darley and Batson 1973;\n  Blackwell and Dubins 1962, Annals of Mathematical Statistics; White 2005; Schoenfield 2014.\n\nSee also: `convergence.md` § The accounting · `philosophical-foundations-stress-test.md` § 4 ·\n`starting-values-and-the-problem-of-the-prior.md` § 6.6 · `moral-convergence-metaethics-and-psychology.md` § 4 ·\n`empirical-deliberation-convergence-evidence.md` § (e) · `the-canon-distillation-pass.md` § 4j.\n"}