Memo 12 — Synthesis: The Interface to Computable Governance

Series: The Landscape of Normative Systems (LNS) Series ID: LNS Memo number: 12 of 12 Primary JD questions: Q4 (the full dependency graph, every missing layer), Q5 (normative feedback loops, synthesized), Q7 (the relationships among politics, law, ethics, religion, culture, economics, markets, technology, organizations, power, legitimacy, authority, trust, incentives), Q8 (what must never enter the kernel, with justification), Q10 (what remains after removing all ideology — the candidate interface). Status of D0: Immutable. Not modified, not criticized, not defended. This memo specifies a candidate interface to the kernel; it does not touch the kernel's internal design.

Scope note. This is the capstone memo. It does not introduce new sources; it integrates the eleven prior memos into a single answer to the program's founding question: what content-neutral structure must any normative system be reduced to before it can be handed to the D0 kernel, and what must be kept out? The memo assembles the complete dependency graph above the kernel (Q4), synthesizes the feedback loops that make that graph a dynamical system rather than a pipeline (Q5), draws the relationship map among the load-bearing abstractions (Q7), states and justifies the exclusion boundary (Q8), and concludes with the invariant that survives the removal of all ideology (Q10) — the candidate interface between political philosophy and Computable Governance. Per M6, the memo holds open that the honest answer may be "the invariant is too thin to be useful," and it states precisely the one open question (the localizability of residue, inherited from Memo 11) on which the whole program turns. All claims carry epistemic tags (§0.4); content is separated into the four standing categories (§0.5).

12.0 Orientation: what a synthesis must and must not do

INFERENCE The temptation of a synthesis is to declare victory — to announce that the kernel exists and here is its interface. That would violate M6 (discovery, not invention) and M3 (architecture from hostile counterexamples). This memo instead does three disciplined things:

1. Assemble the graph and the relationships the survey memos established, so the compiler architect has one map instead of eleven (Q4, Q5, Q7).
2. Subtract everything the series showed to be content-laden, contested, or mutable — the exclusion (Q8).
3. Report the residue — whatever structure survives subtraction — as a candidate interface, explicitly tagged, with the surviving threats attached (Q10).

FACT The residue is a candidate, not a result. Memo 11 established that the ambitious thesis (CG-STRONG) does not survive, and that the modest thesis (CG-WEAK) survives only if the un-representable residue is localizable. This memo therefore cannot certify the interface; it can only specify what the interface would be conditional on localizability, and name the experiment that would settle it.

12.1 Q4 — The complete dependency graph above the kernel

INFERENCE The JD's seed graph is Politics → Constitution → Law → Regulation → Procedure → Job Description → KPI. Memos 01, 02, 07 showed this is a fragment of a much larger multi-rooted directed graph. Below is the assembled graph. Edges are typed by their transformation semantics (the operation that turns the upstream node into the downstream one), because the edge type — not the node label — is what a compiler must implement.

12.1.1 Edge-type legend

12.1.2 The assembled vertical (deepened seed graph)

12.1.3 The layers the seed graph omitted (Q4: "identify every missing layer")

INFERENCE The JD asks explicitly for the missing layers. The survey memos supply at least the following, each absent from the seed chain:

• Constituent power / founding act above the constitution (Memo 01): the graph has a root that is itself not produced by any node in the graph — the bootstrap (Kelsen's Grundnorm / Hart's rule of recognition, Memo 02 §02.10).
• The legitimacy supply, running upward (Memos 03/04/06): power, culture, religion, morality, and trust do not sit in the vertical chain; they feed bindingness into it from the side. This is the single most important structural correction to the seed graph — legitimacy is an input edge, not a layer.
• Adjudication / precedent as a lateral norm-emitter (Memo 02, G10): courts read from and write back to the Legislation/Constitution nodes.
• Interpretation on nearly every edge (Memo 02): every downward edge that crosses under-determined text hides an INTERPRETS step (open texture, Memo 11).
• The enforcement back-edge (Memo 07 §07.9): compliance/audit closes the loop from the leaf back to upper nodes; the seed graph drew only the forward pipeline.
• The role/authority spine (Memo 07 §07.11): a parallel DELEGATES graph of offices and powers runs beside the content graph; a norm is valid only if issued by a node with authority on the spine (Hohfeld second-order relations, Memo 10).
• Cross-source lateral edges (Memos 04/05/06/08): contract, custom, market norm, treaty, and standard are side-roots that inject norms into the vertical at many levels (a contract binds like private legislation; a technical standard is incorporated by reference into regulation).

INFERENCE Structural finding for D1: the "graph" is not a tree and not a pipeline. It is a multi-rooted DAG for content (what the norm says) laid over a separate DAG for authority (who may say it), with feedback back-edges for enforcement and side-inputs for legitimacy. A compiler that models only the seed pipeline will be wrong about validity (misses the authority spine), about meaning (misses interpretation), and about dynamics (misses the back-edges).

12.1.4 The two-graph model (the compiler's real target)

HYPOTHESIS Validity is computable as a join between the content DAG and the authority DAG: a norm is valid iff there exists an authority-path licensing the content-node that produced it. This is one of the few genuinely machine-checkable governance properties the series identified (Memo 10 §10.6, machine-friendly cluster). Falsifiable: if real systems routinely treat norms as valid without any traceable authority-path (pure custom, G07; emergent convention, G14), the join model is incomplete — and indeed those are exactly the author-less classes Memo 10 flagged as not fitting the machine-friendly cluster.

12.2 Q5 — Feedback loops, synthesized

INFERENCE Memo 09 enumerated eighteen normative feedback loops with control-theory classifications. The synthesis point for the interface is not the individual loops but four structural facts that any executable governance layer must respect, because they convert the Q4 DAG into a dynamical system:

1. The graph is cyclic in time even though it is acyclic in authority. Society elects the legislature that makes the law that shapes society that elects the next legislature (Memo 09's Election→Government→Law→Society→Election). The authority DAG is acyclic at any instant; the system is a feedback loop across instants. INFERENCE Therefore D1 must be bitemporal (Memo 09 handoff): every norm carries both when-it-was-enacted and when-it-applies, or loop reasoning is impossible.

2. Legitimacy is a slow feedback variable, not a constant. Compliance feeds legitimacy feeds easier enforcement feeds compliance (Memo 06/09). This loop is bistable (Memo 09): high-trust and low-trust equilibria both self-sustain. INFERENCE The kernel cannot treat legitimacy as a fixed input tag; it is a state variable updated by the system's own behavior. This is the deepest reason legitimacy cannot be internalized (see §12.4).

3. Positive-feedback loops (cascades, bubbles, ratchets, Goodhart) are the instability sources. Memo 09 classified norm-cascades (G15), market bubbles (G05), bureaucratic ratchets (G12), and KPI-surrogation (G08) as amplifying loops. INFERENCE These are exactly the fused/emergent clusters of Memo 10 that the compiler represents worst. The instability lives where the representation is weakest — a coincidence the architect must not ignore.

4. Delay is the primary hazard. Memo 09 (control theory) established that a negative-feedback (stabilizing) loop with sufficient delay oscillates. Governance loops have long delays (elections, litigation, cultural change). INFERENCE An executable governance layer that closes a loop faster than the system was designed for can destabilize a previously stable norm (the automation risk). D2 must model loop delay explicitly, not assume faster is better.

The three coupled clocks (Memo 09 §09.0, restated as an interface constraint):

The kernel must keep these on SEPARATE clocks; conflating them is the commonest modeling error the series found (Memos 06, 09, 11).

12.3 Q7 — The relationship map

INFERENCE The JD lists fifteen abstractions and asks for their relationships as dependency diagrams. The series supports the following map. The organizing insight (from Memos 01, 06, 11) is that the fifteen split into three strata: a substrate of pre-normative social facts, a machinery layer that produces norms, and a product layer of norms and their measurement. Legitimacy, authority, power, and trust are substrate; politics, law, organizations, markets are machinery; regulation, contract, KPI are product.

12.3.1 The core dependency diagram

12.3.2 The load-bearing relationships (stated as directed claims)

• INFERENCE Power ≠ Authority ≠ Legitimacy. Power is capacity to impose (G13); authority is a recognized right to issue norms; legitimacy is the subjects' disposition to treat norms as binding. The series treats conflating these as the primary category error (Memos 01, 06, 11). Power without legitimacy is imposition (G13, fused validity/efficacy); authority is power that has been legitimated.
• INFERENCE Trust → Legitimacy → Authority → (valid) Law. Trust (built by reliable repeated interaction, Memo 05/06) is the substrate resource; it is spent-down or accumulated by the enforcement loop (§12.2). Legitimacy is the stock; trust is partly its flow.
• INFERENCE Culture/Religion/Ethics supply the content of value; they do not sit in the authority DAG. They feed the LEGITIMATES edge and populate the free parameters of norms (Memo 03 §3.6.4). This is why they are excluded from the kernel (§12.4) yet indispensable above it.
• INFERENCE Economics/Markets couple to Politics/Law via incentives in both directions (Memo 05): law shapes markets (property, contract enforcement); markets shape law (public choice, regulatory capture — quarantined as descriptive, Memo 05).
• INFERENCE Technology is special: it can move the boundary of the possible, not just the permitted (Memo 05/10/11, code-as-law G18). It is the one abstraction that can become the enforcement mechanism (fused cluster), which is precisely what the CG program itself does — the self-referential node in the map.

12.4 Q8 — What must never enter the kernel, and why

INFERENCE The series produces a single exclusion principle from which every specific exclusion follows, plus a test the architect can apply mechanically.

12.4.1 The exclusion principle

A concept must stay above the kernel if binding it requires a value choice, a contested truth-value, or a mutation the frozen kernel cannot track.

This is the union of the three independent legs proven for ethics in Memo 03 §3.6, shown for religion in Memo 04 §4.6, and re-derived adversarially in Memo 11: - Non-neutrality: the concept selects a contested position (essentially contested concepts, Memo 03). - Non-verifiability: the kernel cannot mechanically check its truth (Memo 03 §3.4.5; Harnad grounding, Memo 11 §11.15.2). - Non-amendability: the concept legitimately mutates, but the kernel is frozen (Memo 09; Memo 03 §3.6.3).

12.4.2 The specific exclusions (JD Q8 list) and their justification

INFERENCE Every item on the JD's exclusion list fails the principle for a stated reason, not by fiat. The pattern: each is a binding of a free parameter in an otherwise content-neutral shape. "Democracy" binds the aggregation rule of G01; "justice" binds the objective of G08 / the value-standard slot of Memo 03 §3.6.4; "capitalism" binds the property regime of Memo 05. The shape is admissible; the binding is the ideological act that stays above the kernel.

12.4.3 The mechanical test for the architect

12.5 Q10 — What remains after removing all ideology: the candidate interface

INFERENCE Subtract every excluded concept (§12.4) from the entire landscape (Memos 01–08), keep only what survived destructive testing (Memo 10 §10.7) and the adversarial memo (Memo 11 §11.18), and the following residue remains. This is the candidate interface between political philosophy and Computable Governance. It is stated as a specification with each element's provenance and epistemic status.

12.5.1 The invariant structure (the residue)

  THE CANDIDATE INTERFACE  (content-neutral; carries SHAPE, never content)
  ------------------------------------------------------------------------
  I1. NORM-TOKEN                                                    [INFERENCE]
      a typed record: <modality, subject, condition, consequence,
                       provenance, generation_class, temporal-validity>
      - modality: the minimal core from Memo 10 §10.7 —
          Hohfeld 1st-order (right/duty, privilege/no-right)
        + Hohfeld 2nd-order (power/liability, immunity/disability)
        + dyadic conditional deontic  O(x | c)
        + a defeasible-priority marker
        + an explicit supererogation/optional flag
      - provenance: which source authored it (tag, not evaluated)
      - generation_class: one of G01..G18 (Memo 10) + property vector
      - temporal-validity: bitemporal stamp (Memo 09)

  I2. AUTHORITY RELATION                                            [HYPOTHESIS]
      a partial order of offices/powers; norm is WELL-FORMED iff
      licensed by an authority-path (the two-graph join, §12.1.4).
      This is the machine-checkable validity primitive.

  I3. CONFLICT-RESOLUTION ORDER                                     [INFERENCE]
      the lex superior / posterior / specialis lattice (Memo 02) +
      defeasible priority (Memo 10) — a partial order over norms, with
      an explicit UNDECIDED verdict when the order does not decide.

  I4. THREE-VALUED APPLICATION                                      [INFERENCE]
      apply : Norm x Facts -> {satisfied, violated, UNDECIDED/escalate}
      (Memo 11 §11.14/§11.15: penumbra, undecidable, ungrounded cases
       MUST be escalatable, never silently guessed.)

  I5. GROUNDING BOUNDARY (oracle)                                   [INFERENCE]
      thick/evaluative predicates ("reasonable", "cruel") are bound
      OUTSIDE the kernel by a fallible, provenance-tagged oracle with
      an error model (Memo 03 §3.5; Memo 11 §11.15.2).

  I6. MUTATION INTERFACE                                            [INFERENCE]
      typed edits over the norm-set (Memo 09): add/repeal/amend/
      reinterpret/override, non-append-only, branchable, with AGM-style
      revision for non-monotonic update.

  I7. LEGITIMACY as an EXTERNAL STATE VARIABLE                      [OPEN]
      NOT computed by the kernel; supplied and updated from above as a
      slow, possibly-bistable state (§12.2). The kernel reads it as a
      tag; it never derives it. (Open whether even this is honest — see
      §12.6.)
  ------------------------------------------------------------------------
  EXPLICITLY OUTSIDE: all value bindings, all objective functions, all
  substantive predicate extensions, the exception/emergency trigger
  (Schmitt, Memo 11 §11.8), and legitimacy generation.

12.5.2 Why this is the residue and not more

INFERENCE Each element is what every source in Memos 01–08 needed regardless of its ideology: - Every source produces norms with a modality, a subject, a condition — hence I1 (the shape is invariant; Memo 03 §3.3.6, Memo 10 §10.5). - Every source with an author needs a validity test — hence I2/I3 (Memo 02). - Every source has hard cases it cannot self-resolve — hence I4 (Memo 11). - Every source uses thick predicates it grounds socially — hence I5 (Memo 03). - Every source changes over time — hence I6 (Memo 09). - Every source relies on bindingness it does not itself manufacture — hence I7 as an external variable (Memos 06, 11).

INFERENCE And it is not more because everything else the sources carried — which value, which objective, which rights, which faith, which regime — failed the §12.4 admissibility test. The interface is deliberately, provably thin: it is the grammar of norms, not their vocabulary.

12.5.3 Why this is only a candidate (M6 honesty)

OPEN The interface is conditional on the pivotal open question inherited from Memo 11 §11.17: is the un-representable residue localizable or pervasive? I4 (three-valued application) and I5 (grounding oracle) assume the hard cases can be named and escalated rather than being smeared through every case. If Foucault's constitutive-power thesis or the generative residue of specification-gaming (Memo 11 §11.15.3) is correct, then the UNDECIDED bucket is not a small nameable remainder but the majority of real cases, and a kernel that escalates the majority is not governing anything — it is a formalism that has quietly relocated all the work to the oracle. In that case the honest conclusion is M6's acceptable outcome: no useful kernel exists, and the residue I1–I7 is a description of what norms look like, not a machine that can run them.

12.6 The four standing categories

12.6.1 Established consensus (across the series)

• FACT Norms share a describable structure (modality + subject + condition + consequence); this shape is content-neutral (Memos 03, 10).
• FACT Validity, legitimacy, and efficacy are distinct and change on different clocks (Memos 01, 06, 09).
• FACT Substantive value concepts (justice, rights, specific faiths, regime labels) are contested and mutable (Memos 03, 04, 05, 11).
• FACT Formal systems have hard limits, correctly bounded (Memo 11 §11.14).

12.6.2 Competing theories (unadjudicated)

• [competing] Whether legitimacy is externalizable (I7) or is produced only in human judgment/communication (Rawls/Arendt/Luhmann, Memo 11). The interface assumes externalizable and flags the assumption.
• [competing] Whether the right top-level index is the 18 generation classes, the three clusters, or enforcement coupling (Memo 10 §10.8.2).
• [competing] Positivist (validity = pedigree, I2) vs interpretivist (validity involves moral reading, Dworkin) accounts of the validity primitive (Memo 02).

12.6.3 Open questions

• OPEN The pivot: is residue localizable or pervasive? (Memo 11 §11.17; §12.5.3). The go/no-go of the entire program.
• OPEN Can legitimacy (I7) be honestly externalized, or does externalizing it hollow out the kernel?
• OPEN Is the authority-path validity join (I2) adequate for author-less norms (G07/G14), or do they require a separate, non-pedigree validity notion?
• OPEN Is there a sound finite representation of defeasible + contrary-to-duty norms adequate for real governance (Memo 10 §10.7)?
• OPEN Can the exception/emergency trigger be given a decidable boundary, or is it necessarily outside the rules (Schmitt, Memo 11)?

12.6.4 Research opportunities (the six-month D1/D2 program)

• INFERENCE Build I1 first. Implement the norm-token with the minimal modality core and the generation-class + property vector; ingest a real corpus (a body of regulation + SOPs, Memo 07's machine-friendly cluster) and measure coverage.
• INFERENCE Build the two-graph validity join (I2/I3) and test it on the machine-friendly cluster where it should work, then deliberately on custom/ convention where Memo 10 predicts it fails — a direct M3 hostile test.
• INFERENCE Instrument the pivot (the single most important experiment). Take a real domain, run three-valued application (I4), and measure the size and growth of the UNDECIDED bucket as coverage increases. If it shrinks and localizes, CG-WEAK is supported; if it grows or spreads, M6 fires. This operationalizes the whole program's open question into a measurement.
• INFERENCE Ship the destructive-test suite (Memo 10's four modality breakers
• Memo 11's attack checklist) as continuous conformance tests, so every D1 change is re-subjected to M5.

12.7 Handoff to D1/D2 — the consolidated bill

The eleven prior memos converge on the following instructions. This is the permanent upstream reference the JD asked for.

1. INFERENCE Represent the norm as I1 (typed token, minimal modality core: Hohfeld 1st+2nd order + dyadic conditional deontic + defeasible priority + supererogation flag). Do not freeze the bare O/P/F triple; Memo 10 proved it insufficient.
2. INFERENCE Model two graphs, not one (content DAG + authority DAG); compute validity as the join (I2). The seed pipeline Politics→…→KPI is a fragment.
3. INFERENCE Make application three-valued (I4): satisfied / violated / UNDECIDED-escalate. Never silently guess on penumbra, undecidable, or ungrounded inputs.
4. INFERENCE Bind thick predicates and all value parameters OUTSIDE the kernel (I5, §12.4). The kernel carries shape; content enters as provenance-tagged, mutable input.
5. INFERENCE Keep validity, efficacy, legitimacy on separate clocks (§12.2); treat legitimacy as an external state variable (I7), never a kernel output.
6. INFERENCE Make the norm-set bitemporal, branchable, and non-monotonically revisable (I6, Memo 09); design for feedback-loop delay and instability.
7. INFERENCE Declare the expressiveness fragment and stay decidable for core checks; do not overclaim Gödel/Turing (Memo 11 §11.14).
8. INFERENCE Flag the fused cluster (G08/G18) — code-as-law, algorithmic norms — as the self-referential danger zone where the program's own output lives.
9. OPEN → the gate Run the localizability experiment (§12.6.4) before committing to CG-WEAK. If the UNDECIDED bucket does not localize, M6 has fired and the correct decision is to retreat to "no kernel," which D0's methodology explicitly permits.

12.8 Closing statement of the series

INFERENCE The Landscape of Normative Systems, surveyed across eleven memos and synthesized here, yields a narrow but precise result. The vocabulary of governance — democracy, justice, rights, faith, regime, ideology — is above the kernel, provably, by a single stated principle (§12.4). The grammar of governance — a typed norm-token with a Hohfeldian/deontic modality, a two-graph validity join, a conflict order, three-valued application, a grounding boundary, a mutation interface, and an external legitimacy variable — is the candidate interface, and no more (§12.5). OPEN Whether that grammar can actually run a real polity, rather than merely describe one, reduces to a single measurable question — does the un-representable residue localize? — and per M6 the series ends by handing that question to D1/D2 as an experiment, not by assuming its answer. The strongest honest claim the series supports is: if the residue localizes, this interface is what sits between political philosophy and Computable Governance; if it does not, the same interface is the precise specification of what could not be computed — which is itself a result worth having.

— End of Memo 12 and of the LNS series.