← Research Library

R-21 · 27 min · Paper

Seal the Decision, Not the Document

Published Nov 12, 2026

Abstract

Three papers removed three discretions. Part II proved a verdict can be computed with the agent as no input; Part III proved the applicable law can be enumerated with the carrier as no input. Both rest on an assumption neither examines: that a seal made with the institution's key expresses the institution's decision. Custody is not authorization. An officer holding the sealing key can mint an authority the institution never adopted, and every verifier on Earth — including ours — will call it genuine, because cryptographically it is. The seal proves the institution's key spoke; nothing yet proves the institution decided. We close the gap with one reflexive move: sealing is an act, and Part II already specified how an act carries its authorization. An organization's constitution — who may propose, who must consent, what quorum, within what windows — compiles to a sealed authority like any other; the act "seal artifact with digest D as version V" is evaluated under it; and the resulting committed Act, the issuance proof ι, travels with the artifact, verifiable offline by the same verifier that checks everything else. Four further conditions (C9–C12) make the construction sound, each necessary by an attack built from genuine keys alone: the faithless officer, the midnight amendment, the stale constitution, the self-certified genesis. Their joint consequence is officer indifference: whatever any key-holder does, a valid authority implies the institution's own procedure, current at the time, authorized exactly that sealing — with the regress of constitutions terminating not in a key but in a genesis the relying party's pinned root names. Onchain governance, far from being the hard case, is the easy one: a DAO vote is already a machine-witnessed consent with a quorum proof, so binding "proposal N passed" into issuance needs no new cryptography. Hart gave Part II adjudication and Part III recognition; this paper mechanizes his third secondary rule, the rule of change — applied, as it always ultimately is, to the question of how valid law is validly made.

R-19 · Paper

Authorize the Act, Not the Agent →

Part II, which this paper extends: sealing is itself an act, carried and verified the way Part II specifies.

R-20 · Paper

Resolve the Law, Not the Catalogue →

Part III, the paper before this one: the applicable law enumerated with the carrier as no input.

R-18 · Position Paper

What Remains Human →

The Issuer, Escalation Target and Sovereign whose consent the issuance proof records.

P-03 · Proof

The proofs: Issuance →

A sealed authority binds only when the institution, by its own constitution as it stood when the seal was made, authorized exactly that sealing, with every consent from a holder certified for its role, whatever any key-holder signs.

Lean 4 · 17 attack vectors · found 2 gaps in SPEC-015 0.1.0

Update · Nov 12, 2026

Two corrections, from the proof

The paper is unchanged. Mechanizing its decision core in Lean found two gaps in NOMOS-SPEC-015 0.1.0, both closed in 0.2.0. Read §4 and §5 with them:

  • C10 judges currency when the seal was made, as the relying party knows it from its own record, not at the time the issuance proof states. Judged at the proof's own time, a backdated proof reached a superseded constitution. The proof's time must lie within the relying party's bound of the seal's.
  • C9's consents count only from holders certified for the role they claim. Counted by claimed role, the faithless officer could write his own quorum, for a policy or for a friendlier constitution. Officer indifference (§5) holds with both corrections; it is proved in Lean for SPEC-015 0.2.0.
  • The proofs, the 17 vectors and SPEC-015 0.2.0 are at github.com/nomos-spec/spec; the plain-language page is under Research, Proofs, P-03.

Concept

1. Introduction

1.1 The assumption three papers stood on

Every guarantee in this series flows from a seal. Part I compiled an institution's rules into a sealed artifact; Part II let a stranger verify that an act was permitted under it; Part III let a stranger establish which sealed artifacts govern at all. At the bottom of each proof sits the same quiet premise, stated nowhere and used everywhere: the key is the institution. A chain of custody — the key minted in the organization's own HSM, certified under its root, operated with dual control — establishes that whoever sealed could. It establishes nothing about whether the organization decided.

Part II was candid about the same gap one level down. Its §9.2 concedes that certification “bounds which facts a key may state, not whether its statements are true.” The issuance analogue has been hiding in plain sight: custody bounds who can seal, not whether the sealing was validly decided. Part III closed the analogous gap for law — absence of an authority became load-bearing only under signed closure. This paper closes it for the institution itself: a seal becomes load-bearing only under proof of the decision behind it.

1.2 The faithless officer

The attack needs no theft. A chief compliance officer — correctly appointed, correctly holding the sealing key, every certificate genuine — seals version 2.4 of the trading policy with one limit quietly raised. No board saw it. No committee voted. Under Parts I–III the artifact is perfect: the seal verifies, the chain resolves to the institution's own pinned root, resolution lists it as the in-force authority, every agent and every relying party on Earth now enforces it. The institution's authority has been exercised by exactly the thing this series exists to prevent: one party's unilateral discretion, wearing the institution's name.

And the familiar epistemic twist, the confused-deputy signature of this whole series: the faithless officer and the duly authorized one are indistinguishable at the moment of sealing. Both hold the key. Both produce bit-identical artifacts. The difference between them is not in the cryptography at all — it is in a meeting that did or did not happen. Which is precisely why the remedy cannot be more key discipline. It must make the meeting itself verifiable.

1.3 The question, asked from outside

We did not reach this paper only by introspection. Practitioners of onchain governance, on first contact with this work, ask one question almost verbatim: how does a published authority record reflect the organization's legally valid governance decisions — particularly when those decisions are made onchain? The question is better than it first appears, in both directions. It names the real gap (the record–decision link, not the record's integrity). And its hardest-looking case is actually the easiest: an onchain vote is a consent that is already signed, already quorum-counted, already timestamped by a machine every verifier can check. The hard case is the offchain institution, whose decisive event is minutes in a drawer. §6 is honest about that asymmetry; §7 builds on its easy half first.

1.4 Sealing is an act

The move is reflexive, and the series has already paid for all of its parts. Part II defines how one exact action carries its authorization: an Act binds the action's digest to a sealed authority, witnessed facts and named consents, and a relying party verifies it offline. Nothing in that definition requires the action to be a payment. Let the action be: “seal digest D as version V of authority N.” Let the authority it is evaluated under be the organization's own constitution — its bylaws, compiled in the same Studio that compiles everything else: who may propose a sealing, which roles must consent and how many, what windows and quorums apply. The committed Act that results — receipt and all — is the issuance proof ι, published beside the artifact it authorizes, in the detached pattern attestations already use (never under the seal hash, so ι can be attached without re-sealing).

Verification composes: a stranger who today checks seal + chain + resolution adds one step — verify ι with the same SPEC-008 verifier, against the constitution in force at sealing time, resolved by the same supersession machinery as any other authority. Eval all the way down, grounded where everything in this series is grounded: in what the relying party pinned.

1.5 Contributions

  • The custody assumption, named (§1, §3): the key-holder is the last unexamined discretionary party in the construction; the faithless officer defeats Parts I–III with genuine artifacts alone.
  • A model of constitutional issuance (§3): the constitution as a sealed authority; sealing as an Act under it; the issuance proof ι; amendment as supersession authorized under the predecessor; the regress terminating at a pinned genesis.
  • Four further conditions and their necessity (§4): C9 issuance-as-act, C10 constitutional currency, C11 genesis pinning, C12 amendment under the old law — each necessary by an attack from genuine keys alone.
  • Sufficiency (§5): officer indifference — whatever any key-holder does, a valid authority implies the institution's current procedure authorized exactly that sealing — and the containment arithmetic: a stolen sealing key alone mints nothing; compromise is bounded by the constitution's own quorum.
  • A construction (§7) mapped onto what already ships — SPEC-008 consents, passkey-signed Registry review, institution-held keys, detached attestations — and the SPEC-015 delta, with the onchain witness as the first adapter.
  • The honest limits (§6, §9): machine-verifiable procedure is not legal conclusiveness; offchain governance events enter only through witnesses whose truthfulness is assumed, exactly as Part II's A4 assumed it — now at least explicit, scoped and revocable.

1.6 Relation to the series

Part I compiled authority; Part II carried it; Part III found it; this paper legitimates its birth. The three Hart secondary rules Part II claimed for the machine are now each fully mechanized: adjudication (Part II), recognition in both halves (Part II's pedigree, Part III's enumeration), and with this paper change — the rule governing how valid law is validly made, applied where it bites hardest, to the lawmaker's own acts. R-18's roles gain none: every role needed here already exists, which is the point — Issuers, Approvers, Witnesses and the Sovereign, turned on the institution itself. We propose the construction be specified as NOMOS-SPEC-015, Draft until a second implementation, per the house rule that has not been broken yet.

2. Lineage

As in Parts II and III: each tradition solved one fragment and lacked another, and the construction assembles the fragments.

The corporate seal itself — this paper's subject, practiced for seven centuries without cryptography. The common seal of the medieval and early-modern corporation was never a mere stamp: at law, it could be affixed only pursuant to a resolution of the governing body, typically by two named officers in each other's presence, recorded in the minute book. The physical object was kept in a chest precisely because everyone understood the problem this paper formalizes — whoever held the seal could speak as the corporation. The tradition contributes the exact norm we mechanize: a seal is valid per resolution, not per possession. What it lacked was any way for a stranger to check; the minute book stayed in the drawer, so third parties were given the doctrine below instead.

Ultra vires and apparent authority — the common law's forced compromise. Nineteenth-century doctrine held corporate acts beyond granted powers void (ultra vires; Ashbury Railway v. Riche, 1875), which protected shareholders and ruined counterparties, who could not possibly verify a stranger's internal authorization. So the law swung: the indoor management rule (Royal British Bank v. Turquand, 1856) and apparent authority let outsiders rely on an officer's ostensible power, and modern statutes largely abolished ultra vires for third parties. Read as protocol design, this is the key insight of the section: the law chose reliance over validity because verification was impossible. Every doctrine in this paragraph is a workaround for the missing ι. Make the resolution checkable by strangers at near-zero cost, and the dilemma the common law spent 150 years splitting dissolves — counterparties get validity and reliance.

Notaries and apostilles — procedure witnessing as infrastructure. The notarial act does not certify a document's truth; it certifies that a procedure occurred — this person, before me, on this date, executed this instrument — and the apostille chain (Hague 1961) makes that certification recognizable across borders without the verifier contacting the notary. Contribution: the witness-to-procedure role, and the insight that procedure attestations compose across jurisdictions through a chain of recognized authorities — our ι verified under a pinned root is the apostille with the travel time removed. What it lacks: the notary attests execution, never corporate validity; nobody notarizes the quorum.

Company registries — public issuance records, unverifiable contents. Companies House, the Handelsregister, every corporate registry on Earth: filings of resolutions, officer appointments and charters, public precisely so third parties can rely on them. Contribution: the state already believes issuance records belong in public registries — the institutional slot for ι exists. What it lacks: a registry records that a filing was made, not that the meeting it describes occurred or carried; filings are prose, verified by nobody, and late by months.

Kelsen's basic norm — the regress and where it stops. Validity, in Kelsen's account, is conferred by a higher norm: the by-law is valid under the charter, the charter under the statute, the statute under the constitution — and the regress terminates not in another norm but in the Grundnorm, the basic norm presupposed, not proven. This is §3's amendment regress exactly, and it names the honest terminus: a chain of constitutions, each adopted validly under its predecessor, must bottom out in a genesis that is pinned, not proved — the relying party's presupposition, chosen in the open, exactly as Part II's roots and Part III's registrar pins are. The series' epistemology has been Kelsenian from the start; this paper is where it becomes explicit.

DAO timelocks — issuance-as-act, already running in production. The Governor-plus-timelock pattern that dominates onchain governance (Compound and its descendants) enforces precisely C9 without naming it: executable actions can enter the timelock only as the payload of a proposal that passed; the executor keys alone can execute nothing that governance did not queue. Contribution, twofold: proof the construction is practical at scale with adversarial stakes, and the ready-made witness — a passed proposal is a signed, quorum-counted, timestamped consent any verifier can check. What it lacks is everything offchain: the pattern binds code execution to votes, not institutional authority over the world to decisions — which is the bridge ι builds, in both directions.

Six fragments: validity-per-resolution from the seal's own history, the reliance dilemma from corporate doctrine, procedure witnessing from the notariat, the public issuance slot from the registries, the pinned terminus from Kelsen, and a running existence proof from the DAOs. No tradition has all six. The model assembles them.

3. The model

Notation follows Parts II and III. Nothing in this section introduces a new cryptographic object; every part is a Part II object pointed at a new target.

3.1 The constitution

An organization's constitution K is a sealed authority whose subject matter is the organization's own acts of governance. Its rules read facts about a proposed governance act: who proposes (proposer_role), what kind of act (governance_action — seal, amend, appoint, revoke), which authority it touches, its version lineage. Its escalations name the consent requirements — board_majority (role: director, n = ⌈m/2⌉+1), unanimous_members, token_vote — exactly the role_e/n_e structure SPEC-008 escalations already carry. Its data contract declares which facts come from witnesses (wit: the minutes oracle, the chain) and which from the relying record (rp). K is compiled, sealed, versioned, superseded and revoked like any other authority, because it is one.

3.2 The issuance act and its proof

To issue (or amend, or revoke) an authority N at digest D, the organization performs the act

a_issue = { type: 'authority.seal',
            subject: N,
            params: { digest: D, version: V, supersedes: V_prev } }

and carries it as a Part II Act under K: witnessed facts establishing the proposal's provenance, and consents from the constitution's named approvers over exactly this act's digest — the binding that already makes an approval useless for any altered action (Part II C3). The relying party for this Act is whatever system of record publishes the organization's authorities — a SPEC-013 registry, the organization's own registrar log. On AUTHORIZED commit, the receipt-bearing Act is retained as the issuance proof ι(N, V) and published detached beside the artifact, in the attestation pattern: never under the artifact's seal hash, so ι attaches without re-sealing and its absence is detectable without breaking anything.

A stranger's verification of N then composes four checks, all offline, all with the one verifier: the seal (Part I), the issuer chain (SPEC-007), the applicable-set membership (Part III), and now ι — that a committed Act authorizes exactly this digest as exactly this version, AUTHORIZED under the constitution in force at the Act's time, whose own currency is read from the organization's chain by Part III's supersession machinery.

3.3 The regress, and the genesis

K is an authority, so its own adoption needs ι(K, ·) — evaluated under its predecessor K_prev. The regress is real and must terminate honestly. It terminates the way everything in this series terminates: the genesis constitution K₀ is named — by digest — in what the relying party pins (the organization's root certificate carries its genesis digest, or the pin names both), and every later K is reached from K₀ by a chain of issuance proofs, each AUTHORIZED under the one before. Kelsen's structure, machine-checkable: validity flows down the chain; the top is presupposed, openly, by the party choosing to rely — never asserted by the organization about itself. A relying party that pins an organization's root today already presupposes exactly this much; the model makes the presupposition carry the constitution rather than just the key.

Two consequences fall out before any theorem. Amendment is supersession with a proof: K_{i+1} binds only through ι valid under K_i — the rule of change, literally executed. And the sealing key demotes from sovereign to clerk: it still signs, because something must, but a signature without ι is a draft, not an authority. The key's compromise model changes accordingly (§5).

3.4 Where ι lives

The registrar log (Part III) gains nothing structurally: ι is content-addressed and the existing applies/listed entries MAY carry its digest in entry, so the public chain witnesses that an issuance proof existed at listing time, and anyone can fetch and verify the proof itself. A relying party that demands ι (C9) refuses authorities whose entries lack it — which gives organizations a migration path: authorities issued before this paper verify as before, and a relying party raises the bar per scope, when its risk requires, by configuration rather than flag day.

4. Four further conditions, each necessary

The method unchanged: each condition names the simplification it forbids; each necessity argument is an attack from genuine keys and correctly signed artifacts alone, executable as a vector the day a reference implementation exists. The numbering continues the series (Part II established C1–C4, Part III C5–C8).

C9 — Issuance as act. An authority binds a relying party only with an issuance proof: a committed Act, digest-bound to exactly this seal and version, AUTHORIZED under the organization's constitution.

Forbids: custody as authorization — any verifier for which a valid signature from the institution's certified key suffices to bind.

Attack (the faithless officer). As §1.2: the officer seals an unapproved version with the genuine key. Every check in Parts I–III passes. Under C9 the artifact is a draft wearing a signature: no ι exists, because no consents exist, because no decision happened — and the absence is not a suspicious circumstance but a verification failure, detectable offline by anyone.

C10 — Constitutional currency. The issuance Act is evaluated under the constitution in force at the Act's time, established by the organization's own chain (supersession per Part III), never by any party's assertion.

Forbids: evaluating issuance against a cached, superseded or carrier-chosen constitution.

Attack (the stale constitution). The board tightened K in June: sealing the trading policy now needs the risk committee's consent too. The officer assembles ι in November with consents satisfying April's K — all genuine, all correctly signed, honestly AUTHORIZED under the old text. A verifier that takes the constitution the proof presents accepts governance the institution explicitly withdrew. Part II closed this for the authority an Act cites (C1.2, which version); C10 closes it for the constitution an issuance cites — the carrier's freedom to choose versions, dead at the third and last level.

C11 — Genesis pinning. The constitutional regress terminates at a genesis named in what the relying party pins. A chain of constitutions that certifies its own origin establishes nothing.

Forbids: self-certified corporate identity — trusting K₀ because the organization's key signed it.

Attack (the self-certified genesis). An adversary with any key mints a fresh “constitution,” a chain of amendments, and an issuance proof perfectly valid under them — a complete, internally consistent, counterfeit institution. Every signature verifies; the regress checks out; the whole edifice is sound relative to itself. Only C11 distinguishes it from the real one: its genesis digest appears in no relying party's pin. The attack is Part II's §4.2 (“whose root”) reborn one level up, and the containment is the same: legitimacy is conferred by the relying party's choice, in the open, never by the presenter's self-description.

C12 — Amendment under the old law. A constitution's successor binds only through an issuance proof AUTHORIZED under its predecessor. There is no other door.

Forbids: constitutional replacement by key — treating a newer sealed K as current because it is newer and genuinely signed.

Attack (the midnight amendment). The officer who cannot satisfy K's consent requirements for the policy simply seals a new constitution first — one where the CCO alone suffices — then issues the policy validly under it. Without C12, both artifacts are genuine and the coup is complete and silent. Under C12 the new K binds nobody: its own ι fails under the old K, and every proof built on it fails transitively. The rule of change protects itself first — which is exactly Hart's observation about why the secondary rules are the ones a system cannot leave informal.

Remark on independence. The four attacks separate pairwise, as in Part III: the faithless officer defeats C9 with every constitution current and the genesis pinned; the stale constitution defeats C10 with ι required and genuine; the counterfeit institution defeats C11 with issuance and amendment discipline fully intact inside its own tree; the midnight amendment defeats C12 while C9–C11 all hold. And all four leave C1–C8 untouched: C1–C4 discipline the act, C5–C8 the law, C9–C12 the lawmaker.

5. Sufficiency: officer indifference

5.1 The property

Officer indifference. Whatever any key-holder signs, assembles or withholds, an authority that binds a relying party implies: the organization's constitution, current at issuance time on the organization's own chain, reaching the relying party's pinned genesis by valid amendments, AUTHORIZED exactly this seal of exactly this content — with every consent the constitution requires, given by distinct, certified holders of the named roles, over this issuance and no other.

The argument composes with the prior parts rather than repeating them. By C9, binding requires ι, and ι is a Part II Act — so carrier indifference applies to it wholesale: its facts are witnessed, its consents are digest-bound (no approval transplants to an altered seal), its silence is INCOMPLETE, never acquiescence. By C10 the constitution evaluating it is the current one; by C12 every constitution on the chain was adopted under its predecessor; by C11 the chain ends in the relying party's presupposition, not the presenter's. The officer's residual powers are exactly two: propose a sealing (anyone can), and refuse to operate the key (a denial of service the constitution's own escalation and the organization's key ceremony handle, not a corruption). The last discretionary party in the construction is reduced to what the first paper reduced the agent to: a proposer.

5.2 What one stolen key can no longer do

The compromise arithmetic inverts, and this is the operational payoff.

  • The sealing key alone: nothing binds. Before this paper, the sealing key was the institution; its theft was catastrophic and its revocation a race. Under C9 a thief mints drafts. The key demotes from the crown jewel to one signature among several — still worth protecting (it can propose convincingly), no longer worth a kingdom.
  • One director's consent key: a vote, not a veto-proof coup. Compromise yields one consent toward a quorum the constitution set. The blast radius of any single key is now written in the organization's own K — an institution chooses its compromise tolerance when it chooses its quorums, and the choice is legible to every counterparty reading its constitution.
  • A quorum of consent keys: the honest limit. An adversary holding enough keys to satisfy K is the institution for verification purposes — as a forged quorum of directors is at law. No cryptographic construction distinguishes a captured majority from a majority; what the construction guarantees is that the capture must be majority-sized, simultaneous, and leaves ι — a permanent, signed, public record of exactly which keys consented to the coup. Detection and attribution, where prevention is definitionally impossible: the CT bargain, for corporate capture.
  • The recovery story is constitutional, not cryptographic. A compromised K is superseded by K′ under K's own amendment rule (C12), consents from the uncompromised quorum; revocations (SPEC-006, chain revocations) travel as they already do. The organization heals by governing, which is the correct primitive.

5.3 Reproducibility, extended to the decision

Part II's Theorem 10 let a third party recompute a verdict; Part III extended the record to the enumeration. ι extends it to the decision: an auditor — a court, a regulator, an acquirer's diligence team — holding the artifact, ι, the constitution chain and the pins recomputes not merely “was this permitted” but “did the institution, by its own rules as they then stood, decide this” — offline, years later, without the organization's cooperation. The question that today costs a discovery motion and a deposition becomes a hash comparison. §6 is precise about which word in that sentence — decide — the machine does and does not establish.

6. What issuance proof must not claim

Three boundaries, stated as design obligations. Each marks a place where claiming more would make the construction less trustworthy, not more.

Procedure, not legal conclusiveness. ι proves the constitution's procedure was satisfied: these roles, these consents, this quorum, this window. It does not prove the meeting was lawful — a court may later void the vote for fraud, duress, a conflicted director, a defective notice period expressed nowhere in the compiled K. The right analogy is the notary's: attestation of execution, not adjudication of validity. What changes is the default and the burden: today a counterparty can verify nothing and relies on apparent authority; under ι the procedural record is complete, signed and public, so the residual legal questions are litigated over facts everyone already agrees on. The machine narrows the dispute to what genuinely requires a court; it must never pretend to be one.

The offchain witness is Part II's A4, and we say so. A board that votes in a room enters the construction only through a witness — a corporate secretary's key, certified for minutes:*, stating that resolution R passed. Certification bounds what the secretary may attest, not whether the attestation is true; a faithless secretary is a faithless witness, Part II §9.2's open problem exactly, relocated — not solved, but transformed: the unverifiable thing shrinks from “what did the institution decide?” (unbounded, private, contestable) to “did this named, certified, revocable officer truthfully attest one datable event?” (scoped, signed, permanently on the record, legally actionable as falsification). Onchain governance is the case where even this residue vanishes — the chain is a witness nobody appointed and nobody can suborn — which is why §7 builds the onchain adapter first, and why the question arrived from that community before we wrote this down.

Legitimacy of the constitution itself. ι verifies issuance under K; it is silent on whether K is just, or even sensible. An organization may adopt a constitution concentrating everything in one founder's consent — ι will faithfully prove that concentration was exercised. As with Part III's registrars: the construction's contribution to legitimacy is exposure, not judgment. A counterparty can now read the governance it is trusting — quorums, windows, who can amend what — as a compiled, verifiable object rather than a PDF summary. Pricing that governance is the counterparty's job, and markets are better courts of reasonableness than verifiers are. The machine holds the mirror; it must not hold the gavel.

7. The construction, and how much already ships

As with Part III, the table is the quiet argument: no new cryptographic object anywhere in the model, and most rows are in production for other purposes.

Model object (§3)Shipping todayThe SPEC-015 delta
The constitution KStudio compiles policies with escalations naming roles and consent counts; org roles, dual control, approver rostersA governance-act vocabulary (governance_action, proposer/subject facts) — a context vocabulary, like SPEC-014's
The issuance ActSPEC-008 Acts end to end: digest-bound consents, witnesses, passkey approvers (WebAuthn-ES256), accept()The action type authority.seal and its params schema
Consents over the sealingThe Registry's review-and-release: passkey signatures over the evidence fingerprint, dual control, typed-name approval putting a version in forceRe-emitted as SPEC-008 consents inside ι, so the trail a stranger can verify is the trail the Registry already produces
ι detached beside the artifactAttestations: detached signatures outside the seal hash, travelling with the fileι as a second detached object; entry digest on the registrar log's listing
Constitutional currency (C10)Listing versions + supersession entries (SPEC-014)Nothing new: K is an authority on the same chain
Genesis pinning (C11)SPEC-007 roots, pinned by relying partiesThe genesis digest carried in/beside the root certificate
Amendment under old law (C12)VERSION_TRANSITIONS lifecycle + proposals needing reviewer + approverThe transition gated on ι valid under K_prev
The onchain witnessSPEC-007 witness certificates with scoped claimsOne adapter: a witness attesting chain:<id> proposal <n> passed, from a light-client proof or an attested RPC quorum

Why onchain first. The adapter for a Governor-style vote is a weekend: the vote's outcome is public, signed by the chain's own consensus, and maps one-to-one onto a consent (role: token_holder_majority, n: 1 — the chain aggregates the quorum itself). A DAO issuing a sealed authority with ι bound to its proposal gets something it has never had: offchain reliance on onchain governance with no trusted bridge — a bank, a counterparty, a court can verify that the DAO's own process authorized this exact policy, using nothing but public data and the one verifier. That is the precise answer to the question §1.3 quoted, and the first institution to want this construction is therefore likely to arrive from that side — a reason to treat the asker as a design partner, not an objection.

The order of building mirrors the house pattern: vocabulary + vectors first (the four attacks of §4 as fixtures), the pure evaluation of ι through the existing verifier (it is only an Act — most of the “implementation” is a fixture format), the Registry emitting ι on its existing approval flow, the onchain witness adapter, and last the offchain minutes witness, because it is the assumption-bearing part and deserves the longest soak.

8. What changes

Five consequences, in the series' discipline: each follows from the construction, none from enthusiasm.

The key stops being the institution. Fifty years of PKI practice equates an organization with its private keys, which is why key ceremonies feel like coronations and key theft like regicide. Under C9 the equation breaks: keys sign, constitutions authorize. Institutional identity relocates from a secret (fragile, stealable, binary) to a procedure (public, quorum-bearing, self-amending) — which is where the law always located it; the cryptography has finally caught up with the company law.

“Who sealed this?” becomes “who decided this?” — for strangers. Inside organizations that distinction always existed; outside, it was invisible, which is why apparent authority exists. ι makes the governance pedigree of every authority a public, verifiable property. Due diligence on a counterparty's policies, a regulator's question about who approved a model's deployment rules, a court's question about whether the board knew — each becomes a verification, not an investigation.

Insider risk at the root collapses to quorum size. The catastrophic single insider — the admin, the CCO, the founder with the key — is replaced by arithmetic the organization itself chose and published. Compromise cost scales with the constitution's own consent requirements, and — the inversion worth noticing — an institution's security posture becomes legible in its governance document, where counterparties can price it.

Onchain and offchain institutions converge on one verification. A DAO's vote and a board's resolution become the same object at the point of reliance: a consent inside ι, checked by the same verifier. The bridge both worlds keep attempting — legal wrappers for DAOs, tokenized boards for companies — gets its missing primitive: not a new entity type, but a common proof format for “the institution decided.”

Hart's architecture closes. Adjudication (Part II), recognition in both halves (Parts II and III), and now change, applied reflexively to the lawmaker. The historical thesis extends one final step and stays falsifiable: institutions will delegate consequential authority to software at scale only once the institution's own decisions — not merely its artifacts — are verifiable by the strangers asked to rely on them. If agent-scale delegation thrives on bare key custody, the thesis fails. If every bare-custody deployment eventually meets its faithless officer, it stands. The common law has run this experiment once already, for 150 years, and its answer — the whole doctrine of apparent authority — is what verification-by-strangers now makes unnecessary.

9. Limitations and open problems

9.1 The minutes witness. Offchain governance enters through an attesting officer, and A4 is load-bearing there as it always was. The mitigations are real — scoped certification, revocation, a permanent signed record, and the possibility of two independent witnesses to one meeting (secretary plus external auditor) — but the gap between “attested” and “true” is not closed, only shrunk and priced. How institutions should select and bond issuance witnesses is open, and is the same open problem as Part II §9.2, which we decline to pretend is new.

9.2 Compiling constitutions. Bylaws are prose with the usual pathologies — ambiguity, cross-reference, the clause nobody has read since 1987. The extraction pipeline that compiles law compiles bylaws too, but the stakes of a miscompiled quorum are constitutional by definition. The review bar for K must exceed the bar for ordinary authorities; what that means procedurally (dual compilation? mandatory counsel sign-off as a consent on K's own ι?) is open.

9.3 Privacy of governance. ι names who consented. Boards sometimes have reasons — safety, regulation, acrimony — to prove a quorum without naming the quorum. Threshold signatures prove k-of-n without identifying the k; selective disclosure (Part II §9.3's trade) applies to consents as to facts. Reproducibility versus confidentiality, once more; the right default may differ for public authorities and private companies.

9.4 The genesis ceremony. C11 moves the unprovable moment to the founding: who attests that K₀ is the organization's constitution? For a registered company the answer has existed for two centuries — the state's registry — and a registrar attestation on the genesis is the natural bridge; for everything else (DAOs, consortia, informal bodies) genesis is a social fact the pin simply records. Naming this honestly beats engineering around it.

9.5 Jurisdictional recognition. Whether a court will treat ι as evidence of due authorization — or eventually as presumptive of it, the way statutes came to presume electronic signatures — is a question for legislators and litigators, not verifiers. The construction's job is to make the record so complete that the legal question becomes narrow. The advocacy is someone else's paper.

9.6 Independence of evidence. Nothing in this paper has a second implementation, and most of it has no first: as of this draft, the model is specified and the four attacks exist as executable vectors (fixtures/spec-015: 14 vectors, 28 CI assertions covering necessity, independence and shuffle determinism) against a reference decision core — but the end-to-end path, a real ι as a committed SPEC-008 Act emitted by the Registry's approval flow, is unbuilt. The gate sequence is the house one: vectors, reference implementation, the Registry emitting ι on its real approval flow, one organization — ideally the onchain asker — issuing a real authority under a real constitution, then the independent implementation that ends Draft.

10. Conclusion

This series began by removing the most conspicuous discretion — the agent's — and has ended at the least conspicuous: the officer whose key custody silently stood in for an institution's will. Part II made the verdict carrier-indifferent; Part III made the law carrier-indifferent; this paper makes the authority itself officer-indifferent. What remains in anyone's hands, at every level, is the same thing: a proposal. The act is proposed by the agent; the law is proposed by the registrar within its certified scope; the authority is proposed by whoever drafts it. Nothing binds until the procedure that strangers can check has run.

The corporate seal spent seven hundred years locked in a chest, valid only per resolution, unverifiable beyond the room. The common law built an entire doctrine — apparent authority — as a workaround for that unverifiability, deliberately choosing to let organizations be bound by officers who exceeded their mandates, because the alternative was commerce grinding on minute-book inspections. That trade was correct for its costs. The costs have changed. A resolution can now travel with the seal it authorized, checkable in milliseconds by anyone, forever. The chest is open; the question is only how long institutions keep paying apparent-authority prices for decisions they could prove.

Intended for the series alongside Parts I–III, with the construction specified as NOMOS-SPEC-015 once the vectors exist. We invite the same three things as always — independent implementations, attempts to break the obligations, corrections — and one more, specific to this paper: an organization, onchain or off, willing to issue its first authority under its own constitution, in public.

Cite this

This is a field in formation; this page will be revised. Cite the version date below, not the day you happen to be reading this.

Suggested citation

SafeHaven (2026). Seal the Decision, Not the Document. Computable Authority Research Library, version of 2026-11-12. https://www.computableauthority.com/research/seal-the-decision-not-the-document

BibTeX

@misc{computableauthority-seal-the-decision-not-the-document,
  author       = {SafeHaven},
  title        = {Seal the Decision, Not the Document},
  howpublished = {Computable Authority Research Library},
  year         = {2026},
  note         = {Version of 2026-11-12},
  url          = {https://www.computableauthority.com/research/seal-the-decision-not-the-document},
}

Full citation guidance →