# Moral Convergence Analysis: Deriving the Invariant Set **Status:** Research input to Kernel v0.1. The output of this document is Article IV of the kernel. **Question:** Across human moral traditions — religious, philosophical, legal, and as measured empirically — what is the *small* set of norms that (a) nearly everyone converges on, and (b) can be enforced by procedure rather than by agreeing on a worldview? That set, and only that set, gets entrenched in the kernel. --- ## 1. Method: convergence, not consensus, and three filters We are not seeking moral truth and not seeking unanimity. We are seeking the **overlapping consensus** (Rawls's term, used here in its engineering sense): norms that distinct traditions arrive at *from incompatible premises*. A Buddhist, a Kantian, a Confucian, and a contractarian disagree about why you may not torment the weakest member of the group; they agree that you may not. Independent derivation from incompatible premises is the moral analogue of the comparative method in `03-synthesis.md`: it is evidence the norm tracks something real about cooperation, not something parochial about one culture. A candidate norm enters the invariant set only if it passes all three filters: - **Filter A — Convergence:** attested in substantially all of the surveyed traditions (allowing for differences of scope and emphasis). - **Filter B — Proceduralizability:** expressible as a constraint on *rules and procedures* rather than on beliefs or private conduct. "Honor your parents" converges widely but is not a constraint a constitution's amendment pipeline can enforce. "No rule may single out a named person for punishment" is. - **Filter C — Testability:** violations must be detectable — mechanically where possible, by adversarial scenario otherwise. An invariant that cannot fail a test is decoration (failure mode F11). Everything that passes A but fails B or C is **deliberately pushed to userland** (§5). This is the load-bearing design decision: the kernel entrenches the convergent *meta*-morality and stays silent on substantive morality, which is exactly where traditions diverge and where parameterization belongs. --- ## 2. Survey: four independent lines of evidence ### 2.1 Religious and wisdom traditions | Tradition | Reciprocity formulation | Other kernel-relevant norms | |---|---|---| | Judaism | "What is hateful to you, do not do to your fellow" (Hillel, Shabbat 31a) | Equal justice for stranger and citizen (Lev 24:22); judicial impartiality (Deut 16:19); public reading of the law (Deut 31:10–13) | | Christianity | "Do unto others as you would have them do unto you" (Mt 7:12) | The least of these as the moral measure (Mt 25:40); equal moral standing of persons | | Islam | "None of you believes until he wishes for his brother what he wishes for himself" (an-Nawawi 13) | No bearer of burdens bears another's burden (Q 35:18 — individual, non-collective responsibility); consultation (*shūrā*, Q 42:38); written, witnessed contracts (Q 2:282) | | Hinduism | "One should never do to another what one regards as injurious to oneself — this is the sum of dharma" (Mahābhārata, Anuśāsana Parva 113.8) | *Ahiṃsā* (non-harm); duties of rulers toward the weakest (*Arthaśāstra*'s protection obligations, whatever its realism elsewhere) | | Buddhism | "Comparing oneself to others... one should neither kill nor cause others to kill" (Dhammapada 129–130) | Five precepts as a minimal harm-floor; right speech (truthfulness as social infrastructure) | | Confucianism | "What you do not wish for yourself, do not impose on others" (Analects 15.24) | *Ren* (humaneness); rectification of names (Analects 13.3 — **rules must mean what they say**: a 2,500-year-old statement of the legibility requirement); remonstrance (the duty and right to object to authority) | | Daoism | Non-imposition; governing a great state "like cooking a small fish" | Anti-overreach: a standing caution against rule-maximalism — support for a *minimal* kernel | | Ubuntu (Southern African) | "Umuntu ngumuntu ngabantu" — a person is a person through other persons | Dignity as relational and non-forfeitable; restorative rather than purely punitive response; consensus-seeking deliberation (*indaba*) | | Indigenous American traditions (e.g., Haudenosaunee Great Law) | Mutual obligation across the confederacy | Seventh-generation consideration; structured voice for constituent nations; documented influence on federal design | Two observations. First, the Golden Rule's negative ("silver") formulation — *do not impose what you would not accept* — is more universal than the positive one and is the directly proceduralizable form: it becomes the **generality requirement** (a rule's author must be willing to live under it; therefore rules may not name their targets). Second, several traditions independently flag *legibility of rules* as a moral matter, not a technical one (public reading of the law; rectification of names; written contracts) — which licenses entrenching transparency, not merely preferring it. ### 2.2 Secular philosophical traditions | Tradition | Convergent contribution | |---|---| | Kant | Universalizability (act only on maxims you could will as universal law) → generality requirement; persons as ends → dignity floor, no member is purely instrumental to the group | | Utilitarianism (Bentham/Mill) | "Each to count for one, nobody for more than one" (Bentham via Mill) → equal suffrage; Mill's harm principle → bounds on what rules may reach | | Contractarianism (Rawls) | Veil of ignorance → design rules as if you don't know which member you'll be; the **difference principle** → evaluate arrangements by the position of the worst-off. This is the direct philosophical ancestor of the project's empathy metric, and we adopt it explicitly as the scoring rule, not as an invariant (see §5.3) | | Scanlon (contractualism) | Rules must be justifiable to each person affected — no one's standing may be simply overridden → voice and hearing rights | | Natural law / Fuller | Fuller's eight ways law fails (*The Morality of Law*, 1964): rules must be general, public, prospective, clear, non-contradictory, possible to follow, stable, and congruent with official action. This is the closest thing the literature has to a **prior art kernel invariant list**, and §4 maps to it | | Capability approach (Sen/Nussbaum) | Moral assessment by what the worst-positioned can actually *do*, not formal entitlements → informs the empathy metric's measurement, guards against dead-letter rights (F11) | ### 2.3 Modern rights instruments (negotiated cross-cultural convergence) The UDHR (1948) is the largest empirical experiment in moral convergence ever run: drafters from Confucian (P.C. Chang), Thomist (Maritain, adjacent), Islamic, liberal, and socialist traditions producing one text. Maritain's famous report of the drafting: *"we agree about the rights but on condition that no one asks us why."* That sentence is this document's method stated in 1948. Kernel-relevant articles, each convergent with §2.1–2.2: - Equal dignity and rights (Art. 1); non-discrimination in rights (Art. 2) → **dignity floor, generality** - Recognition as a person before the law; equality before the law (Arts. 6–7) → **equality before the rules** - No arbitrary detention/exile; fair hearing by an impartial tribunal (Arts. 9–10) → **due process before sanction** - No retroactive offenses (Art. 11.2) → **non-retroactivity** - Freedom of opinion/expression and assembly (Arts. 19–20) → **voice** - Right to take part in government; equal suffrage; will of the people as the basis of authority (Art. 21) → **suffrage** - Right to leave any country (Art. 13.2); right to change nationality (15.2) → **exit** The ICCPR's non-derogable core (Art. 4.2: even in declared emergencies, no derogation from the bans on arbitrary killing, torture, slavery, retroactive punishment, or non-recognition as a person) is the international system's own answer to our question "what stays entrenched even under Article VI emergencies" — and our Article IV/VI interaction copies its structure. ### 2.4 Empirical anthropology and game theory - **Curry, Mullins & Whitehouse (2019), "Is It Good to Cooperate?" (*Current Anthropology* 60:1):** coded ethnographies of 60 societies for seven cooperation-derived morals — help kin, help your group, reciprocate, be brave, defer to superiors, divide resources fairly, respect prior possession. The seven were judged morally good in 99.9% of codable observations across all regions, with **zero societies coded as treating any of them as bad**. For the kernel: *reciprocity* and *fairness in division* pass Filter A empirically, not just textually. ("Defer to superiors" also converges — and is exactly the kind of norm we entrench *procedures around* rather than entrench itself.) - **Brown (1991), *Human Universals*:** reciprocity, sanctions for wrongs, distinction of in-group rules, conflict-resolution mechanisms appear in every documented society. - **Game theory:** iterated cooperation is stabilized by reciprocity with forgiveness (Axelrod), by reputation/indirect reciprocity (Nowak & Sigmund), and degraded by unaccountable punishment. **Proportionality of sanction** is not just felt to be fair; disproportionate punishment empirically destabilizes cooperation. **Ostrom's principles** 4–6 (monitoring accountable to members, *graduated* sanctions, cheap local conflict resolution) are the field-observed versions of due process and proportionality in commons that survived for centuries. --- ## 3. Candidate norms through the filters | Candidate | A: Converges? | B: Proceduralizable? | C: Testable? | Verdict | |---|---|---|---|---| | Generality / reciprocity (no self-exempting or person-targeting rules) | ✔ (every tradition, §2.1–2.4) | ✔ constraint on rule *form* | ✔ mechanical (named targets) + scenario (engineered targeting) | **INVARIANT** | | Non-retroactivity | ✔ (UDHR 11.2, ICCPR non-derogable, Fuller, fiqh, common & civil law) | ✔ | ✔ largely mechanical | **INVARIANT** | | Dignity floor / due process before sanction | ✔ (Ubuntu, imago dei, Kant, UDHR 6–10, Ostrom 5–6) | ✔ procedural rights before any sanction; membership not strippable without process | ✔ scenario-testable | **INVARIANT** | | Voice (speak, propose, vote) | ✔ (shūrā, remonstrance, indaba, UDHR 19–21, Scanlon) | ✔ | ✔ mechanical + scenario | **INVARIANT** | | Exit | ✔ in moral texts (UDHR 13.2); ✔ in non-state practice; suppressed by states (a divergence of *incumbent interest*, not of moral judgment — no tradition argues captivity is good for the captive) | ✔ | ✔ | **INVARIANT** | | Proportionality of sanction | ✔ (graduated sanctions in Ostrom; restorative emphasis in Ubuntu; lex talionis as a *cap* in its original context; sentencing principles everywhere) | ✔ as a constraint + interpretive backstop | ◐ scenario-testable, judgment-heavy | **INVARIANT** (with Article V backstop) | | Legibility (public, written, versioned, non-contradictory rules) | ✔ (Deut 31; rectification of names; Q 2:282; Fuller's eight) | ✔ | ✔ highly mechanical | **INVARIANT** | | Honesty/truthfulness in general | ✔ | ✘ private conduct | ✘ | Userland; kernel captures only the *institutional* slice via legibility and written-reasoning duties | | Care for kin / loyalty | ✔ | ✘ | ✘ | Userland | | Property regimes | ◐ "respect prior possession" converges; *what may be possessed and how* diverges wildly | ✘ in general form | ◐ | Userland (a family, a commune, and a DAO need different answers; forcing one is exactly the over-reach a kernel must avoid) | | Substantive equality / distribution | ✘ diverges (this is most of politics) | ✘ | ◐ | Userland — but *measured* by the empathy metric (§5.3) | | Deference to authority | ✔ empirically (Curry) | dangerous to entrench | — | Inverted: the kernel entrenches *procedures that make authority safe to defer to* (terms, recall, interpretation) | | Sanctity / purity norms | ✘ diverges | ✘ | ✘ | Userland | --- ## 4. The invariant set (output → kernel Article IV) Six invariants survive. Stated here in research form; Article IV gives normative text and annotations. Mapping to Fuller's eight desiderata shown for prior-art traceability. **INV-1 — Generality.** Every rule must be general in form and prospective in application: no rule may name, or be constructed to single out, specific members for benefit or burden, and the rule's authors must be subject to it on equal terms. *(Fuller: generality, congruence. Sources: silver rule everywhere; Kant; bills-of-attainder bans. Counters F10.)* **INV-2 — Non-retroactivity.** No act lawful under the rules in force when performed may be sanctioned afterward; no vote, membership, or standing already vested may be invalidated by later rule change. *(Fuller: prospectivity. ICCPR non-derogable core. Counters F10, and snap-rule variants of F2/F8.)* **INV-3 — Dignity floor.** No member's membership, voice, or vote may be suspended or removed except by a process that existed before the conduct at issue, with notice, a hearing, an impartial decider, and proportionate, graduated sanction. No sanction may be collective. *(Ubuntu; Kant; UDHR 6–10; Ostrom 5–6; Q 35:18 on non-collective burden. Counters F4, F10; this is also where the empathy metric bites hardest in testing.)* **INV-4 — Voice.** Every member may speak on, propose, and vote on matters within the body's competence under the rules' uniform procedures; and on kernel-tier matters, every member's vote counts equally. *(Shūrā, remonstrance, UDHR 19–21, Bentham's "each to count for one." Counters F5, F6.)* **INV-5 — Exit.** Every member may leave at any time without sanction, and the right to fork — to take the rules, the public record, and one's lawful entitlements and continue under a new instance — may not be abridged. *(UDHR 13.2/15.2; the entire open-source canon; Hirschman. Counters F9 and is the backstop for all other failures.)* **INV-6 — Legibility.** All rules in force must be written, public, versioned, and non-contradictory; an act of governance not performed and recorded under the published rules is void. Secret rules and secret decisions bind no one. *(Deut 31:10–13; Analects 13.3; Q 2:282; Fuller: publicity, clarity, non-contradiction, congruence. Counters F1, F11, F12.)* Design properties of the set, by construction: - **Small and orthogonal.** Six, each addressing a distinct failure surface; pairwise non-derivable. - **Negative/procedural in form.** Five of six are constraints on what rules and procedures may do; none requires shared belief. This is what makes the same kernel runnable by a family and a DAO. - **Mutually reinforcing under attack.** Remove any one and a documented capture path through the others opens (e.g., without INV-5, INV-4 voice becomes cheap talk; without INV-6, INV-1–3 are unverifiable). The test suite will include a deletion-ablation scenario per invariant to keep this claim honest. - **Non-derogable.** Per the ICCPR Art. 4 pattern, Article VI emergencies cannot touch them, and the amendment pipeline offers no threshold that amends them — the lawful way to live under different invariants is the Article VII fork. ## 5. What is deliberately excluded, and the empathy metric **5.1 Excluded:** substantive distribution, property regimes, speech norms beyond procedural voice, family/role norms, purity norms, and all positive duties of care. Not because they don't matter — because traditions *diverge* there, and entrenching one tradition's answers would make the kernel a faction. They are userland modules: same kernel, different config. **5.2 The boundary rule:** userland modules may add obligations and structures freely but may not contradict an invariant (kernel Article 0 supremacy). A family module may assign chores; it may not strip a child member of voice without process. A DAO module may weight treasury votes by stake; it may not weight kernel-tier votes (INV-4). **5.3 The empathy metric is a *scoring rule*, not an invariant.** Rawls's difference principle and the capability approach enter the system as the first-graded axis of every constitutional test: *how does the worst-off participant fare under stress?* We do not entrench it as text because (Filter B) "arrange everything to the maximum benefit of the least advantaged" is not a procedural constraint — it is a substantive theory, and entrenching it would violate our own exclusion rule. Instead it lives where this project puts its paranoia: in CI. Optimism in the defaults, the difference principle in the tests.