Memo 10 — Computational Taxonomy of Norms

Series: The Landscape of Normative Systems (LNS) Series ID: LNS Memo number: 10 of 12 Primary JD questions: Q1 (complete enumeration of sources of normative authority), Q2 (six-part analysis lens applied across sources), Q3 (the central deliverable — a taxonomy of norms by computational behavior / generation function rather than by academic discipline). Status of D0: Immutable. Not modified, not criticized, not defended in this memo. (The adversarial treatment of the computability assumption is reserved for Memo 11; here we only stage the threats that arise naturally from the taxonomy.)

Scope note. This is a CORE synthesis memo. It does two things the survey memos (01–08) could not do individually. First (Part A), it consolidates the most complete enumeration in the series of every known source of normative authority, giving each a one-line origin+enforcement gloss so that later work has a single master checklist. Second, and centrally (Part B), it re-cuts the entire landscape along a different axis: not by discipline (politics, law, religion, market, society) but by computational behavior — the generation function that produces the norm. The claim under test (M6: discovery not invention) is that a small set of generation-function shapes recurs across every discipline, and that this shape — not the disciplinary label — is what a compiler on top of D0 actually has to represent. For each behavior class we specify a function signature, input set, determinism/nondeterminism, revisability, verifiability, and enforcement coupling, and we name which sources (from Memos 01–08) instantiate it. We then run methodology M5 (destructive testing) against the shared modality vocabulary of §0.6 — Hohfeld's eight relations and the deontic operators O/P/F — using the classical breakers (Ross's paradox, Chisholm's contrary-to-duty paradox, defeasibility, supererogation) and report what survives. A master ASCII table (behavior-class × property) is the memo's single most reusable artifact. All claims carry epistemic tags per §0.4; content is separated into the four standing categories per §0.5.

10.0 Orientation: why re-cut the landscape by behavior

INFERENCE Memos 01–08 are organized by source discipline. That organization is natural for a survey but is the wrong organization for a compiler. A compiler does not care whether a norm "comes from law" or "comes from religion"; it cares about the shape of the process that produced the norm, because that shape determines what the compiler can and cannot do with it: whether the norm is deterministic, whether it can be re-derived, whether it can be checked, whether it can be revised, and how tightly it is bound to an enforcement organ. Two norms from wildly different disciplines — a papal encyclical (Memo 04) and a military standing order (Memo 01/07) — may share a generation function (from-authority/command) and therefore present identical problems and affordances to a compiler. Conversely, two norms from the same discipline — a statute (from-voting/aggregation) and a common-law precedent (from-adjudication) — may present completely different problems despite both being "law" (Memo 02).

HYPOTHESIS The central hypothesis of this memo, inherited from the series and here made testable: the disciplinary label of a norm is presentation (M4); the generation function is closer to the computational object. If true, the taxonomy of Part B, not the enumeration of Part A, is the correct index into the landscape for D1/D2. This is falsifiable: if we find sources whose behavior cannot be located in any generation-function class, or generation classes that collapse into one another under inspection, the hypothesis is weakened. We report both kinds of strain where we find them.

FACT This memo does not resolve the kernel/above-kernel boundary (that is Memo 12) and does not attempt to break the computability assumption itself (that is Memo 11). Its job is to produce the two enumerations and the stress-test result.

A text diagram of the re-cut:

PART A — Q1: Master Enumeration of Sources of Normative Authority

10.1 Method and reading of Part A

INFERENCE Part A is deliberately flat and exhaustive rather than analytic. Depth-per-source is delivered in Memos 01–08 and is not repeated here; the six-part analysis (JD Q2: origin, legitimacy, enforcement, mutation, hierarchy, conflicts, computational implications) is applied in full only to the behavior classes of Part B (10.5) and to a representative subset of sources below, because the six-part frame transfers more cleanly onto behavior than onto discipline — itself a small piece of evidence for the memo's central hypothesis. Here each source gets a one-line origin+enforcement gloss plus a pointer to (i) the memo that treats it in depth and (ii) the dominant generation-function class it instantiates (forward reference to 10.5). Sources marked † are ones the survey memos treat only in passing or not at all; this memo is their primary home, which is why Part A must be more complete than any single survey.

Notation in the tables: Gen = dominant generation-function class (10.5 codes G01–G18); Enf = enforcement coupling (organ / diffuse / self / none); Memo = depth reference.

10.1.1 The enumeration is open, not closed OPEN

OPEN No enumeration of normative sources can be proven complete. The list below is the most complete in the series and is intended to saturate the ordinary catalog, but M6 forbids claiming closure. Any source discovered later must be locatable in the behavior taxonomy (10.5) even if it is absent from this source list; that is the fallback the memo builds in. If a future source cannot be placed in 10.5, that is a falsification signal for the central hypothesis, not merely a gap in Part A.

10.2 Established-consensus sources (the standard catalog)

These are the sources every survey memo and the JD itself already recognize. FACT that each of the following is standardly treated as a source of norms in the relevant scholarly field; the analysis of each is contested, the existence as a source is not.

10.3 Extended-catalog sources (the JD's "expand much further")

INFERENCE The JD explicitly demands the catalog be pushed past the standard list. The following are genuine, distinct sources of ought-statements that the survey memos treat lightly or not at all. Each is real in the sense that it demonstrably produces norms with which subjects comply and for deviation from which there are consequences. † marks sources for which this memo is the primary home.

OPEN The extended catalog can be continued indefinitely (dietary rules, hygiene norms, dress codes, gaming/e-sports meta-rules, artistic-genre conventions, scientific-citation norms, open-source licensing norms, DAO on-chain governance, house rules, monastic rules, etc.). Per 10.1.1 the memo does not claim closure; it claims that every additional source found so far lands in one of the eighteen behavior classes of 10.5. INFERENCE The recurrence of the same Gen codes down both tables above is the first visible evidence for the central hypothesis: dozens of disciplinarily distinct sources collapse onto ~18 generation shapes.

10.4 Reading Part A: three structural observations

INFERENCE (1) Enforcement coupling clusters. Reading the Enf column, sources fall into four coupling regimes — dedicated organ (state, court, agency, board, referee), diffuse social sanction (culture, etiquette, market reputation), pure self-sanction (personal commitment, ethics, natural law), and automatic execution (code-as-law, algorithmic norms). This four-way split is orthogonal to discipline and reappears as a first-class property column in the master table (10.6).

INFERENCE (2) Most sources are Gen-plural. Almost every source lists a primary and one or more secondary generation classes. This is expected and important: a single source (e.g., religion) can posit content by revelation (G02), transmit it by tradition (G07), and adjudicate disputes about it (G10). A compiler must therefore attach the generation class to the individual norm, not to the source. HYPOTHESIS Generation class is a property of a norm-token, not of a source-type; this is testable by finding any single source whose every norm shares exactly one generation class — the memo predicts such pure sources are rare and small (e.g., a pure lottery mechanism).

INFERENCE (3) The automatic-enforcement sources are new. Code-as-law, algorithmic norms, and platform ToS (the † technology cluster) are the sources least treated by the classical survey and the most consequential for D1/D2, because in them the enforcement mechanism and the norm are the same object (the running code). This collapses the validity/efficacy distinction the whole series relies on (§0.6) and is flagged for Memo 11 and handed to D1/D2 (10.8).

PART B — Q3: Taxonomy of Norms by Computational Behavior

10.5 The generation-function classes (G01–G18)

INFERENCE This is the memo's central deliverable. We classify norms not by where they come from but by the shape of the function that produces them. For each class we give a uniform record:

Type notation: Agents = set of participating parties; Prefs = their revealed preferences/votes; Text = an authoritative source text; Facts = world-state; Precdt = a corpus of prior decisions; Obj = an objective function; Env = a selection environment; ⟶ = "produces". Norm is the shared object of §0.6 (modality, subject, condition, consequence).

G01 — from-voting / aggregation

• SIGNATURE aggregate : Agents × Prefs × Rule ⟶ Norm
• INPUT SET an enfranchised agent set, their expressed preferences, and a fixed aggregation rule (majority, supermajority, weighted).
• DETERMINISM deterministic given the rule and the ballot set; but the outcome as a function of preferences is subject to Arrow / Gibbard–Satterthwaite limits (Memo 01 §2) — FACT no aggregation rule satisfies all fairness desiderata simultaneously; agenda and path can change the result.
• REVISABILITY high and scheduled (next vote); the rule itself is meta-revisable.
• VERIFIABILITY high: procedure is auditable (who voted, count, rule applied).
• ENF-COUPLING organ (the enacting body + downstream apparatus).
• INSTANTIATED legislation (G01 primary), referenda, most parliamentary output, committee decisions, DAO on-chain votes. INFERENCE The cleanest class for a compiler: explicit inputs, explicit rule, auditable trace.

G02 — from-revelation

• SIGNATURE reveal : Source_transcendent ⟶ Text ; interpret : Text × Tradition ⟶ Norm
• INPUT SET an authoritative, typically closed corpus (Text) plus an interpretive tradition/priesthood. The positing act is non-repeatable and non-auditable by outsiders (Memo 04).
• DETERMINISM the base text is fixed (read-only root); the interpretation layer is nondeterministic and school-dependent.
• REVISABILITY base: effectively immutable / very low (schism forks rather than edits); interpretive layer: moderate (responsa, ijtihad, councils).
• VERIFIABILITY low at the root (revelation is not third-party checkable, tied to M1); moderate in the interpretive layer (against text + method).
• ENF-COUPLING diffuse (conscience, community) with organ variants (canon law courts, historically state).
• INSTANTIATED religion (primary), ritual/taboo, natural law (secondary — a "revelation of reason"). INFERENCE Computationally = immutable root-of-trust + mutable interpretive overlay.

G03 — from-contract / agreement

• SIGNATURE agree : Agents × Offer × Acceptance ⟶ Norm(binding on the parties)
• INPUT SET two or more consenting parties, an offer, acceptance, (often) consideration and capacity. Norm scope is local to the parties.
• DETERMINISM deterministic given valid formation; content is whatever the parties wrote (private legislation).
• REVISABILITY by mutual amendment or termination; unilateral revision is a breach.
• VERIFIABILITY high: formation and terms are documentary and court-testable.
• ENF-COUPLING organ (courts) + relational/reputational sanction.
• INSTANTIATED contracts, treaties (states as parties, Memo 08), constitutions qua ratification, oaths/vows, sports' constitutive rules (agreement to be bound), platform ToS (adhesion contract). INFERENCE Bilateral/multilateral scope makes this the natural model for scoped norm namespaces in D1.

G04 — from-consensus (deliberative / standardization)

• SIGNATURE converge : Agents × Deliberation ⟶ Norm (no formal vote; "rough consensus")
• INPUT SET a participating body, an issue, iterated deliberation until no sustained objection remains ("rough consensus and running code," Memo 08).
• DETERMINISM nondeterministic; path- and participation-dependent; no crisp decision rule (distinguishes it sharply from G01).
• REVISABILITY moderate; superseded by a new consensus / new standard version.
• VERIFIABILITY moderate: the artifact is public but "consensus reached" is a judgment call, not a count.
• ENF-COUPLING diffuse (cost of non-interoperation, conformance testing) + self.
• INSTANTIATED technical standards / protocols (ISO, IETF, W3C), accounting standards, some tribal council decisions, treaty drafting. INFERENCE The absence of a decision rule is the compiler's problem here: correctness is not mechanically checkable.

G05 — from-market-equilibrium

• SIGNATURE clear : Agents × Supply × Demand ⟶ Price/Norm (emergent regularity)
• INPUT SET many self-interested agents, exchange, prices/信号 signals; no legislator. The "norm" is an emergent regularity (trade usage, credit terms) (Memo 05).
• DETERMINISM stochastic/emergent; equilibrium may be multiple, unstable, or absent; sensitive to information (Hayek, Memo 05/11).
• REVISABILITY continuous and decentralized (reprices constantly).
• VERIFIABILITY low as a norm (there is no enactment record); observable only as aggregate behavior.
• ENF-COUPLING diffuse (profit/loss, reputation, refusal to deal).
• INSTANTIATED market norms (primary), lex mercatoria, social conventions with a payoff structure. INFERENCE No enactment event and no author — a hard case for any representation that assumes a norm has a text.

G06 — from-authority / command

• SIGNATURE command : Authority × Directive ⟶ Norm (imperative, immediate)
• INPUT SET an agent with recognized authority over subjects, and a directive. Validity = the commander's authority + (sometimes) the order's lawfulness.
• DETERMINISM deterministic content (the order is the norm); legitimacy supplied from outside the act.
• REVISABILITY immediate by the same or higher authority (countermand).
• VERIFIABILITY high as to content and issuance; the authority is validated elsewhere (chain of command, office).
• ENF-COUPLING organ (hierarchy, discipline, court-martial).
• INSTANTIATED political authority/decree, military doctrine & orders, family/ parental authority, platform ToS (unilateral), sumptuary edicts, executive regulation. INFERENCE The single "hole" is the manifestly-unlawful-order exception (Memo 07) — a decidable-scope check the compiler must model.

G07 — from-tradition

• SIGNATURE transmit : Practice_t × Time ⟶ Norm_{t+1} (bindingness from pastness)
• INPUT SET a long-standing practice plus the belief that its longevity confers authority ("we have always done X").
• DETERMINISM deterministic in content, but the content is discovered, not posited; boundaries are fuzzy.
• REVISABILITY very slow; changes by drift, desuetude, or reinterpretation (Memo 09), rarely by edit.
• VERIFIABILITY low: no enactment record; existence is evidenced by practice + attestation.
• ENF-COUPLING diffuse (expectation) with organ recognition (courts adopting custom).
• INSTANTIATED custom/usage, tribal & indigenous law, kinship rules, honor codes, linguistic norms, diplomatic protocol, criminal/underworld codes. INFERENCE Shares the "no author, no text" difficulty with G05 and G14.

G08 — from-optimization

• SIGNATURE optimize : Obj × Constraints × Data ⟶ Norm (argmax/argmin)
• INPUT SET an explicit objective function, constraints, and data; the norm is whatever the optimizer outputs as best.
• DETERMINISM deterministic given (Obj, data, solver); but choice of Obj is a value act imported from above (Memo 03), and the optimizer is subject to specification-gaming (Goodhart; Memo 07/11).
• REVISABILITY high (re-run on new data/objective); can change silently.
• VERIFIABILITY the computation is checkable; the appropriateness of the objective is not mechanically checkable.
• ENF-COUPLING organ, frequently automatic (the optimizer also acts).
• INSTANTIATED algorithmic/automated norms (ranking, scoring, dynamic pricing), parts of regulatory cost-benefit rulemaking, code-as-law (secondary). INFERENCE The generation class that most tempts the program to smuggle value into the kernel via a hard-coded Obj; must be resisted (Memo 03 §3.6).

G09 — from-evolutionary-adaptation / selection

• SIGNATURE select : Variants × Env × Retention ⟶ surviving Norm-set
• INPUT SET a population of norm-variants, a selection environment, and a retention/transmission mechanism (variation–selection–retention, Memo 09).
• DETERMINISM stochastic; no designer; outcome is fitness-relative to a possibly non-stationary environment.
• REVISABILITY continuous but blind; no goal, only differential survival.
• VERIFIABILITY low; "why this norm survived" is a post-hoc reconstruction (HYPOTHESIS status of memetic explanations, Memo 09).
• ENF-COUPLING none intrinsic; enforcement is itself a selected trait.
• INSTANTIATED the macro dynamics of custom, convention, and market norms over long horizons; no source uses it as a primary, deliberate generator (which is why it is rarely a primary code in Part A). INFERENCE Important for Memo 09/12 dynamics; a warning that some norm-sets have no author to query.

G10 — from-adjudication / precedent

• SIGNATURE adjudicate : Dispute × Norms × Facts ⟶ Ruling ; extract : Ruling ⟶ Norm(precedent)
• INPUT SET a concrete dispute, applicable norms, found facts, and an authorized adjudicator; the ruling both resolves the case and (in precedent systems) emits a new norm (Memo 02).
• DETERMINISM nondeterministic at the margin (open texture, discretion, hard cases — Hart/Dworkin, Memo 11); deterministic in clear cases.
• REVISABILITY by distinguishing, overruling, or legislative override (Memo 09).
• VERIFIABILITY the reasoning is published and checkable for consistency; the rule extracted (ratio vs dictum) is itself interpretively contested.
• ENF-COUPLING organ (courts + enforcement apparatus).
• INSTANTIATED judicial precedent (primary), sports referees (instant micro- adjudication), religious responsa (adjudication within G02). INFERENCE The "norm as a side effect of resolving a case" shape is distinctive: generation and application are the same act.

G11 — from-negotiation / bargaining

• SIGNATURE bargain : Agents × Threats × Concessions ⟶ Norm (a point on the bargaining frontier)
• INPUT SET parties with divergent interests, outside options (BATNA), and relative bargaining power; the norm is the settlement reached.
• DETERMINISM nondeterministic; power- and sequence-dependent; distinct from G03 (contract records a bargain; G11 is the process that finds the terms) and from G04 (consensus seeks agreement, bargaining trades concessions).
• REVISABILITY by re-negotiation when power/outside-options shift.
• VERIFIABILITY low as to process (power is not documented); the result becomes a G03 artifact.
• ENF-COUPLING varies; often self/reputational until reduced to contract.
• INSTANTIATED treaty negotiation (Memo 08), collective bargaining (guild/union), plea bargaining, diplomatic settlement. INFERENCE Power as an input means the norm's content encodes a power ratio — the Foucauldian point (Memo 11) in computational form.

G12 — from-delegation

• SIGNATURE delegate : Principal × Grant(scope) ⟶ Agent-authority ; Agent ⟶ Norm(within scope)
• INPUT SET a principal with authority, an explicit grant defining scope, and an agent who then produces subordinate norms (Memo 07).
• DETERMINISM deterministic as to scope check (ultra vires is decidable against the grant); nondeterministic as to content within scope.
• REVISABILITY the grant is revocable/amendable by the principal; issued sub-norms fall if the grant falls.
• VERIFIABILITY high on the scope dimension (was the sub-norm within delegated authority?); this is one of the most compiler-friendly checks in the taxonomy.
• ENF-COUPLING organ (the delegating hierarchy).
• INSTANTIATED administrative regulation, organizational governance/policy, education, guild rules, netiquette (moderator authority), medical protocols. INFERENCE The scope predicate is a clean, decidable validity test — a primitive Memo 12 should carry.

G13 — from-force / imposition

• SIGNATURE impose : Coercer × Directive ⟶ Norm (validity = capacity to coerce, legitimacy claimed later or never)
• INPUT SET a party with coercive capacity and a directive; consent and prior authority are absent (this is what distinguishes it from G06, where authority is recognized).
• DETERMINISM deterministic content; efficacy tracks coercive capacity, not legitimacy.
• REVISABILITY persists while coercion persists; collapses when it fails.
• VERIFIABILITY content observable; there is no legitimating record to check (that is the point).
• ENF-COUPLING organ = the coercive apparatus itself; enforcement and norm are nearly fused.
• INSTANTIATED conquest/occupation regimes, coup-imposed orders, extortion codes (overlaps criminal codes), martial law at genesis (the Schmittian exception, Memo 11). INFERENCE The class where efficacy is high but legitimacy is null — a stress case for any representation that conflates the three (§0.6).

G14 — from-convention / emergence (coordination equilibrium)

• SIGNATURE coordinate : Agents × RepeatedGame ⟶ Norm (a self-enforcing equilibrium)
• INPUT SET many agents with aligned interest in coordinating, and a history that selects one equilibrium among several (Lewis conventions; Memo 06).
• DETERMINISM stochastic selection among multiple equilibria; once selected, self-stabilizing (deviation is individually costly).
• REVISABILITY low absent a coordinated jump; can flip in a cascade (Memo 09).
• VERIFIABILITY low: no author, no text; existence inferred from behavior + expectations.
• ENF-COUPLING none external — self-enforcing by payoff structure (the purest case of enforcement without an organ).
• INSTANTIATED social conventions (driving side, language), etiquette, culture, fashion (with G15), money-as-convention. INFERENCE Self-enforcement means the "enforcement mechanism" field can be empty yet compliance high — breaks the assumption that every norm needs an enforcer.

G15 — from-imitation / cascade

• SIGNATURE imitate : Observed-behavior × Threshold ⟶ adopted Norm (informational/reputational cascade)
• INPUT SET observable others' behavior and an agent's adoption threshold; norms spread by copying, not by decision (Granovetter thresholds, Memo 06/09).
• DETERMINISM stochastic and highly path-dependent; small early differences → large divergence (tipping).
• REVISABILITY fast and reversible (reverse cascades); volatile.
• VERIFIABILITY none as a norm; measurable only as adoption curves.
• ENF-COUPLING diffuse (status, being current) or none.
• INSTANTIATED fashion/aesthetic norms, norm cascades (Finnemore–Sikkink), culture transmission (with G07/G14). INFERENCE The most volatile generator; any D1 representation must treat these norms as stateful and time-varying, not static rules (Memo 06 handoff).

G16 — from-internalized-intuition (folk moral sense)

• SIGNATURE intuit : Situation × internalized dispositions ⟶ Norm-verdict (pre-reflective)
• INPUT SET a moral situation and an agent's internalized dispositions (upbringing, culture, possibly innate moral psychology); output is a felt ought, prior to justification (Memo 03).
• DETERMINISM fast, automatic, but inter-agent variable; not a stable function across a population.
• REVISABILITY slow (moral development, reform); resistant to argument.
• VERIFIABILITY none: introspective, not third-party checkable; no truth-value the kernel can read (Memo 03 §3.4.5).
• ENF-COUPLING self (guilt/conscience) + diffuse (blame, reputation).
• INSTANTIATED folk morality, much of custom's felt bindingness, honor's internal side. INFERENCE The generator with the least machine-legibility; its outputs enter the system only after being articulated (G17) or codified.

G17 — from-reasoned-justification / expertise (epistemic authority)

• SIGNATURE justify : Premises × Method × Warrant ⟶ Norm (with a reason-trace)
• INPUT SET premises (including at least one normative premise — the is/ought barrier, Memo 03 §3.5), a method of reasoning or a body of expertise, and a warrant claim ("because the evidence/argument shows…").
• DETERMINISM deterministic relative to accepted premises and method; but premises and method are contested (metaethics; scientific underdetermination).
• REVISABILITY by better argument/evidence (in principle continuous; in practice slow — professions and paradigms resist).
• VERIFIABILITY the reasoning is checkable for validity; the soundness depends on contested premises.
• ENF-COUPLING self + diffuse (peer review, argument) with organ variants (licensing boards, IRBs).
• INSTANTIATED systematized ethics, professional & medical standards, science- as-norm-source, accounting standards, natural law, self-imposed commitments (reasoned self-binding). INFERENCE The class that produces a reason-trace — attractive to a compiler because the justification is explicit, dangerous because the premises smuggle value.

G18 — from-automated-execution (code / architecture)

• SIGNATURE embed : Norm ⟶ Architecture ; run : Architecture × Input ⟶ enforced outcome
• INPUT SET a norm already compiled into a system's architecture such that the regulated behavior is physically/logically constrained, not merely prescribed (Lessig's "code is law," Memo 05/11).
• DETERMINISM deterministic execution (the code runs); the norm and its enforcement are the same object.
• REVISABILITY by editing/redeploying code; can change silently and instantly for all subjects at once.
• VERIFIABILITY the code is inspectable in principle; opacity, proprietary closure, and Rice's theorem (Memo 11 §11.14) limit it in practice.
• ENF-COUPLING automatic — enforcement is not a separate stage; violation is often made impossible rather than punished.
• INSTANTIATED code-as-law, smart contracts, DRM, platform architecture, drilled military procedure as "human code." INFERENCE This is the class closest to what the CG program itself produces (D2). It collapses validity/efficacy into one object (§10.4 obs. 3) and is therefore both the target and the sharpest self-referential risk for the program.

10.5.19 A note on saturation (per M6)

OPEN At least one further degenerate generator exists — from-random / lottery / sortition (draw : Candidates × RandomSource ⟶ Norm/appointment, deterministic process / nondeterministic output, high verifiability of the draw, used in jury selection, ancient Athenian sortition, and some allocation rules). No enumerated source uses it as a primary generator, so it is logged here rather than numbered, and it stands as evidence that the class list is not closed (M6). INFERENCE Its existence does not weaken the central hypothesis: it too is a generation shape, locatable on the same property axes as G01–G18.

10.6 Master table: generation-class × computational property

INFERENCE This is the memo's single most reusable artifact for D1/D2. Columns: Det (D=deterministic, N=nondeterministic, S=stochastic); Rev (revisability speed L/M/H); Ver (third-party verifiability of correct production, L/M/H); Auth? (is there an identifiable author to query? Y/N); Text? (is there an authoritative text/record? Y/N); Enf (organ/diffuse/self/auto/none); Val≠Eff? (does the class keep validity and efficacy distinct, or fuse them?).

• * D given a fixed rule + ballots, but Arrow/G-S limit outcome-as-fn-of-prefs.
• ** revelation has no third-party-auditable author (M1 boundary, Memo 04).
• *** self-enforcing by payoff structure — no external enforcer needed (G14).
• **** intuition yields no checkable truth-value; code is inspectable in principle but Rice-limited in practice (Memo 11 §11.14).

INFERENCE Reading the table yields the memo's operative findings:

1. A machine-friendly cluster — G01, G03, G06, G12 (and G10 in clear cases): identifiable author, authoritative text, high verifiability, validity distinct from efficacy, organ enforcement. These are what a compiler can represent most faithfully today.
2. A hard cluster — G05, G07, G14, G15, G16 (and G09): no author, no text, low verifiability, blurred validity/efficacy, diffuse/absent enforcement. These are the emergent/tacit norms (Memos 05/06/09) that resist the "norm has a text" assumption; they must be represented statefully and probabilistically, not as rules.
3. A fused cluster — G08, G18 (and G13): enforcement and norm are the same object; validity and efficacy collapse. This is where the CG program's own output lives and where its self-referential risks concentrate (Memo 11).

HYPOTHESIS The three clusters are the real top-level partition the compiler needs — more useful than the 18 fine classes — and correspond to three different representation strategies (rule-object, stochastic-state-object, executing-artifact). Falsifiable: if a source's norms cannot be assigned to exactly one cluster, or if the representation strategy for a cluster fails on a member, the partition is wrong.

10.7 Destructive test of the shared modality vocabulary (M5)

INFERENCE §0.6 provisionally adopts two vocabularies for representing a norm's modality: Hohfeld's eight jural relations (right/duty, privilege/no-right, power/liability, immunity/disability) and the deontic operators O/P/F. Per M5 we now try to break them with the classical breakers. The question is not "are they elegant?" but "what survives contact with the hard cases, and what must D1 add?"

10.7.1 Ross's paradox (deontic O)

• FACT In Standard Deontic Logic (SDL), O(a) entails O(a ∨ b) (from a ⊢ a∨b and the inheritance rule). So "you ought to mail the letter" entails "you ought to mail the letter or burn it" — intuitively wrong.
• INFERENCE Verdict: the closure of obligation under logical consequence is the culprit. It does not destroy O/P/F but shows SDL over-generates. Survives with restriction: D1 must not treat obligations as closed under arbitrary disjunction-introduction; obligations attach to specified actions, not to their logical weakenings.

10.7.2 Chisholm's contrary-to-duty paradox (CTD)

• FACT The set {O(help); O(if help then tell); if ¬help then O(¬tell); ¬help} is jointly consistent and non-redundant in natural reading, yet SDL cannot represent it without inconsistency or redundancy.
• INFERENCE Verdict: flat O/P/F cannot represent what one ought to do given that one has already violated a primary duty — the essential structure of remedies, penalties, and "second-best" governance (which is most of operative law and compliance, Memo 07). Requires extension: D1 needs dyadic / conditional obligation O(x | c) and an explicit ordering of primary vs contrary-to-duty norms. This is a first-class requirement, not a nicety.

10.7.3 Defeasibility

• INFERENCE Real norms are defeasible: "keep promises," "birds fly," carry open-ended exception sets (Memo 03 §3.7). Monotonic deontic logic cannot withdraw a conclusion when an exception is learned. Requires extension: D1 needs a non-monotonic / defeasible layer with explicit priority relations among norms (connects to Memo 09 AGM revision). Hohfeld and O/P/F describe the modality but not the defeasibility; they must be embedded in a defeasible reasoner.

10.7.4 Supererogation and the deontic gap

• INFERENCE Acts beyond duty (praiseworthy but not obligatory) fall outside the O/P/F partition, which assumes every act is obligatory, permitted, or forbidden (Memo 03 §3.7). The tripartition is exhaustive by construction and therefore cannot name the supererogatory. Requires extension or explicit exclusion: D1 must either add a modality ("optional-but-meritorious") or explicitly declare supererogation out of scope. Silent omission would misclassify it as merely "permitted," losing information.

10.7.5 What survives — the Hohfeld verdict

• INFERENCE Hohfeld's relations survive well as a static description of a norm's jural position, and better than O/P/F on one axis: the power/liability and immunity/disability pairs correctly capture second-order norms (the ability to change norms), which O/P/F cannot express at all. This matters because much of governance is meta-norms (Hart's secondary rules, Memo 02; delegation G12). But Hohfeld is static: it says nothing about conditions, defeasibility, or CTD structure.
• INFERENCE Net result of the M5 test:

INFERENCE Conclusion of §10.7: the §0.6 vocabulary is retained but proven insufficient by itself. The minimal adequate modality core for D1 is Hohfeld (first- and second-order) + dyadic conditional deontic operators + a defeasible priority layer + an explicit stance on supererogation. This is a concrete, falsifiable specification produced by destructive testing, exactly as M5 requires, and it is handed to Memo 12 and D1.

10.8 Four standing categories

10.8.1 Established consensus

• FACT The sources of Part A are all recognized as norm-sources in their fields; their existence is uncontested even where their analysis is.
• FACT Aggregation (G01) faces Arrow / Gibbard–Satterthwaite limits; SDL suffers Ross and Chisholm; Hohfeld's relations are a standard analytic vocabulary.
• INFERENCE Broad agreement that some norms have identifiable authors and texts (G01/G03/G06/G12) and others are emergent and author-less (G05/G07/G14).

10.8.2 Competing theories

• OPEN /competing Whether generation class is a property of a norm-token or a source-type (§10.4 obs. 2) — the memo takes the token side as HYPOTHESIS.
• [competing] Whether the right top-level partition is the 18 classes, the three clusters (§10.6), or the enforcement-coupling four-way split (§10.4). Presented without adjudication; Memo 12 must choose.
• [competing] Whether emergent norms (G05/G07/G14) are "really" norms or merely regularities — a live dispute inherited from Memo 06.

10.8.3 Open questions

• OPEN Is the class list closed? (No; §10.5.19 shows it is not.) What is the minimal generating set of shapes?
• OPEN Can every norm be assigned a single dominant generation class, or are some irreducibly multi-class at the token level?
• OPEN Does the three-cluster partition (§10.6) survive contact with real corpora, or do sources straddle clusters?
• OPEN Is there a sound, finite representation of defeasible + contrary-to-duty norms adequate for real governance (from §10.7)? Unresolved across the series.

10.8.4 Research opportunities

• INFERENCE Build the generation-class tag as a mandatory field on every D1 norm-token, with the §10.6 property vector attached; measure whether real corpora populate all 18 classes and whether the three clusters hold.
• INFERENCE Implement the minimal modality core of §10.7 and run the four breakers as a regression test suite — any D1 norm-logic must pass Ross, Chisholm, a defeasibility case, and a supererogation case, or explicitly declare the case out of scope. This operationalizes M5 as CI.
• INFERENCE Prototype three representation strategies (rule-object / stochastic-state-object / executing-artifact) for the three clusters and find the hostile counterexamples where each fails (M3).

10.9 Handoff to D1/D2

Concrete computational implications for the compiler architects:

1. INFERENCE Index the landscape by generation class, not discipline. Attach a generation_class ∈ {G01…G18, +degenerate} tag to every norm-token; the disciplinary source is metadata/provenance, not the primary key (central hypothesis, §10.0).
2. INFERENCE Carry the §10.6 property vector (Det, Rev, Ver, Auth?, Text?, Enf, Val≠Eff?) per token; it tells the compiler what operations are safe (re-derivation, checking, revision, querying an author).
3. INFERENCE Adopt the three-cluster representation strategy as the top-level design split: rule-objects (machine-friendly cluster), stochastic-state-objects (hard/emergent cluster), executing-artifacts (fused cluster). Flag straddlers (§10.8.3).
4. INFERENCE Author-less norms are first-class. G05/G07/G09/G14/G15 have no author and no text; D1 must not assume a norm can be "looked up" from an authoritative record. These need stateful, time-varying, probabilistic representations (Memo 06/09 concur).
5. INFERENCE Validity/efficacy fusion is a special type. For G08/G13/G18, the §0.6 separation collapses; mark these tokens explicitly so the kernel does not apply validity checks that presuppose a separate enforcement stage.
6. INFERENCE Adopt the minimal modality core from §10.7: Hohfeld (1st + 2nd order) + dyadic conditional deontic O(x|c) + a defeasible priority layer + an explicit supererogation stance. The bare O/P/F triple is insufficient and must not be frozen alone.
7. INFERENCE Ship the four-breaker regression suite (Ross, Chisholm, defeasibility, supererogation) as a mandatory conformance test for any D1 norm-logic (§10.8.4).
8. INFERENCE → Memo 12 The taxonomy is the index into the interface. Memo 12 must decide which generation classes' shape (not content) the content-neutral kernel carries; the machine-friendly cluster is the least-risky starting point, the fused cluster the most self-referentially dangerous.

One-line summary: eighteen recurring generation-function shapes (in three clusters) — not the disciplinary labels — are the real index into the normative landscape; the shared modality vocabulary survives destructive testing only after being extended with conditional and defeasible structure.