Digital Constitutional Systems

A constitution is not distinguished from a statute by what it says. It is distinguished by what it does to the people who can change it: amendment procedures, entrenched provisions, judicial review, separation of the power to write a rule from the power to apply it. The framing under test is that code can carry that function — that a protocol's upgrade procedure, a smart contract's immutability, or an executable encoding of statute can constrain rule-changers the way a written constitution is supposed to.

The evidence base is unusually good for a governance subject, because the artefacts publish their own arithmetic: Tezos records every amendment vote to the roll, Ethereum's 2016 fork was decided by a poll still on chain, Canada publishes every algorithmic impact assessment its departments file, and four national courts have ruled on what a coded administrative decision must disclose. So the framing can be tested against outturns — and the outturn is lopsided: coded systems have executed constitutional content reliably and have never yet bound the people able to rewrite them.

One case cuts the other way, and this brief puts it up front rather than in a footnote. In 2017 a Bitcoin upgrade agreed by parties holding the overwhelming majority of hashing power was abandoned because the users and node operators who held no formal lever refused to run it. That is a rule-changer being stopped. Whether it counts as code doing the stopping is the argument section 2 has to win or lose.

Established The one time an encoded constitution faced a genuine crisis, the decision was made off chain by a handful of people in four days. On 17 June 2016 an attacker drained 3.6 million ETH from The DAO, roughly a third of the ether committed to it. The DAO's code contained a 28-day holding period that bought time; it contained no procedure for what to do next. Established A non-binding “carbonvote” ran on 16 July 2016: of 82,054,716 ETH then in existence, 4,542,416 voted — 5.5% of supply, about 87% in favour (3,964,516) against 13% (577,899), with roughly a quarter of the affirmative weight from a single address. The fork executed four days later, on 20 July 2016, implemented by client developers holding no formal amendment authority; opponents kept running the unforked chain, which became Ethereum Classic. An irreversible partition of a polity, decided outside every stated procedure, in under a week.

Established The chain that exists to entrench immutability had its history rewritten three times in four weeks. In August 2020 Ethereum Classic was 51%-attacked three times: 3,693 blocks reorganised on 1 August, roughly 4,000 in early August, and more than 7,000 — about two days of mining — on 29 August, by attackers who bought hashpower on the open market. Frontier The price moved less than 4%, which is either evidence the market never believed the constitutional commitment or evidence it prices something else; the record does not separate those.

Established In the same period a legislature encoded a factual claim about code into statute, and the claim was false. Arizona wrote at A.R.S. §44-7061(E)(1) that blockchain technology “IS IMMUTABLE AND AUDITABLE AND PROVIDES AN UNCENSORED TRUTH”. Frontier But the statute is weaker evidence than it looks, and the correction runs against this brief's argument. The operative provision, §44-7061(C), says only that a contract “MAY NOT BE DENIED LEGAL EFFECT... SOLELY BECAUSE THAT CONTRACT CONTAINS A SMART CONTRACT TERM”, largely restating existing electronic-transactions law; and Svikhart, in Stanford Law Review Online, concludes the provisions survive federal ESIGN preemption “for now”, precisely because they accord blockchain records equal rather than greater legal status, his “awkward legal limbo” describing blockchain records generally pending congressional clarification. Arizona is a legislature saying something false inside a provision that is legally unremarkable and probably valid.

Established Tezos is the strongest case for the framing and deserves its due before the objection. Its self-amendment procedure — five sequential periods, a 5% proposal floor, an 80% supermajority and a floating quorum, specified in section 7 — has run continuously since 2019, and per the project's own documentation 21 amendments have been adopted between Athens in May 2019 and Ushuaia on 30 June 2026, including replacement of the consensus algorithm and a change of monetary policy. No fork, no crisis, no off-chain rescue. Established The problem is the electorate it runs on. Only delegates — bakers — may vote, with weight equal to staking balance including tez delegated to them. A 2021 report by Cortier, Gaudry and Glondu of CNRS, Inria and the Université de Lorraine, prepared during a collaboration with the principal Tezos development company and so a finding against its sponsor's interest, records the top baker at about 12% of voting power and the top three at about 25%, all votes published in cleartext, and one proposal carried at 99.61% with 117 opposing rolls in which the votes of the 71 largest stakeholders were individually deducible. Its structural point is exact: “it is not because one trusts some entity to bake on one's behalf, that one trusts it for voting.” Frontier Participation sits at 54–58% of that already-narrow set, and the same core company submitted both Athens variants. The procedure works and the polity is small and captured; averaging those into a verdict destroys the information.

Established On concentration, two good measurements disagree because they measure different things, and both are true. Zukowski's August 2026 study in Frontiers in Blockchain audits 52 token protocols on live governance data from March–May 2026, stripping out protocol-controlled addresses. Top-1,000 concentration spans HHI 0.005 for Polkadot to 0.199 for Livepeer, median about 0.04; prior studies inflated HHI by a median factor of 2.3× and up to about 18× by counting treasuries, staking contracts and exchange custody as holdings; launch allocation is uninformative about eventual governance power (r = 0.09, p = 0.55, N = 50); and the robust adverse finding is narrower — delegation amplifies concentration in 13 of 18 protocols, by 2.45× to 25.65×. Frontier Against that, Chainalysis — a vendor selling analytics into this market, though the finding embarrasses its customers — reported in June 2022 that across ten major DAOs less than 1% of holders control 90% of voting power, and that proposal creation needs 0.1–1% of supply and passage 1–4%, so between one in ten thousand and one in thirty thousand holders can unilaterally carry a proposal. Established Zukowski measures holdings net of custodial artefacts; Chainalysis measures threshold accessibility. Both hold — and the second is the constitutionally relevant one, because constitutions are about who can change the rules, not who owns what.

Established Two live cases show the amendment rule performing exactly as written and producing an outcome nobody calls legitimate. In Compound Proposal 289 on 28 July 2024, a group calling itself the Golden Boys accumulated COMP on the open market and carried a proposal moving 499,000 COMP, about $24 million of treasury, into a yield product they controlled; the margin was 682,191 to 633,636, on a third attempt, and one multisig member said publicly he “didn't know this was a vote happening”. Established In Arbitrum's AIP-1 in spring 2023 the Foundation put 750 million ARB, about $1bn, to a ratification vote and began spending and transferring the tokens to itself before the vote closed, while it headed to overwhelming failure. In both cases the rule was satisfied and the rule was not the constraint.

Established Every effective constitutional constraint on an automated system in the record was imposed by an ordinary court or statutory regulator applying pre-digital law. The District Court of The Hague held on 5 February 2020 (ECLI:NL:RBDHA:2020:1878) that Article 65 of the SUWI Act and Chapter 5a of the SUWI Decree were incompatible with Article 8(2) ECHR because the legislation was “insufficiently clear and verifiable”, and ordered the State to stop using the SyRI risk-indication system — while declining to order disclosure of the models or destruction of the data. Established France's Conseil constitutionnel, in décision no. 2018-765 DC of 12 June 2018, permitted purely algorithmic individual administrative decisions on four conditions: disclosure of the algorithm's main characteristics on request, with no algorithmic decision at all where disclosure would breach a protected secret; an administrative appeal whose body must not rely exclusively on the algorithm; judicial review with power to compel disclosure; and no sole determination by sensitive data. Self-revising algorithms are barred outright as the exclusive basis of a decision. Established Italy's Consiglio di Stato, in sentenza 2270/2019 of 8 April 2019, held that an algorithmic teacher-mobility assignment is an atto amministrativo informatico subject to Article 97 of the Constitution, must be “fully knowable” as to authorship, method, data and priorities, and must be open to the administrative judge's full cognition.

Established The two largest remedies are also analogue. In Prygodicz v Commonwealth (No 2) [2021] FCA 634 Australia's Federal Court approved a settlement covering approximately 648,000 group members; the Commonwealth conceded it “did not have a proper legal basis to raise, demand or recover asserted debts based on income averaging from ATO data”, withdrew about $1.763 billion in debts, promised about $751 million in refunds and paid $112 million inclusive of costs, Justice Murphy calling Robodebt “a shameful chapter... and a massive failure of public administration”. Established The Dutch data protection authority fined its own government €2.75 million for using nationality as a risk indicator in automated childcare-benefit screening — processing it called “unlawful, discriminatory and improper”, against a population containing roughly 1.4 million registered dual nationals — and €3.7 million for the FSV fraud blacklist, which held 270,000 people, had no statutory basis and was applied using “nationality and physical appearance”. Not one of these outcomes was produced by code; every one came from an instrument written on paper decades before the system existed — Article 8 ECHR of 1950, Article 97 of the Italian Constitution of 1947, an Australian social-security statute.

Established Code's resistance to amendment has done constitutional work exactly once, and as a limit on the state rather than a source of rule. In Van Loon v Department of the Treasury, No. 23-50669 (5th Cir., 26 November 2024), the court held Tornado Cash's immutable smart contracts are not “property” under IEEPA because “they are not capable of being owned” and “cannot be altered, removed, or controlled” by anyone including their creators. Established That holding has since had its practical effect: OFAC removed Tornado Cash from the SDN list on 21 March 2025, delisting 44 associated addresses. Ungovernability constrained the sanctioning power and constrained nobody inside the protocol. A rule nobody can change is not a constitution; it is a fact about the world that constitutions have to accommodate.

Frontier The strongest counter-case is Bitcoin's SegWit2x, and leaving it out would be the first thing a hostile reader found. In 2017 an agreement among exchanges, custodians and mining operations — parties holding the overwhelming majority of hashing power — committed to a hard fork doubling the block size, and it was cancelled on 8 November 2017 for lack of consensus. The actors holding every formal lever the system has were stopped by node operators and users holding none. Frontier This brief's judgement is that it counts, and that it does not count as the framing needs it to. It counts because a digital polity refused an amendment its most powerful participants had agreed, and because the veto ran to a class with no stake weight, no hashpower and no vote — the opposite of every concentration finding above. Established It does not rescue the framing because Bitcoin has no amendment procedure at all: nothing specified who ratifies, nothing counted the refusal, no rule was applied. What operated was a convention enforced by exit — the economic nodes' ability to run different software and take the ticker with them. Frontier And the power is one-directional: it can block a change, never compel one, and nothing constrains the blockers. The honest score is four for the courts, one for the users, and zero for the code — and the user win was the absence of a rule rather than the operation of one.

Frontier The live positions are incompatible, and the first frontier question is which the evidence has already closed. The strong code-is-law thesis — a protocol's rules as self-executing and thereby self-legitimating — is the one the record closed hardest against: it failed the first time it was tested. Frontier A weaker position holds that code is a regulator alongside law, markets and norms; nothing here refutes it and nothing here makes it constitutional. Speculative A third treats digital constitutionalism as an ideology rather than an instrument, with private ordering acquiring constitutional weight only through the guiding role of state law. Speculative A fourth, Sirota's, separates constitution from constitutionalism: a protocol can have a document of governance while lacking the structural restraints and accountability to the governed that make it constitutionalist. The Aragon and Optimism outturns fit the third and fourth alike and cannot distinguish them.

Frontier Does an amendment rule exist anywhere that binds the people who can execute the amendment? This is the field's central unanswered question and it is empirically answerable. The candidate mechanisms are thin: entrenched clauses with no on-chain enforcement, multi-house designs like Optimism's Token House and Citizens' House, time-locks, and legal wrappers giving a court jurisdiction over developers. Speculative No test case has been run in which an encoded entrenchment clause was invoked against a core development team and held, so the field's central claim rests on artefacts never stressed in the direction that matters.

Frontier Is a coded rule determinate enough to be law? Witt, Huggins, Governatori and Buckley ran a two-week study in which legally trained coders converted Australian Commonwealth copyright provisions into executable form. Agreeing the “key legal terms, or atoms” in advance substantially raised rule similarity, but divergent interpretive choices persisted, attributed to statutory-interpretation complexity, inconsistent granularity and coding-language limits. Established Two competent lawyers encoding the same section produce different code, which is the whole problem in miniature: an authoritative coded rule would make the coder the interpreter. Frontier Small n, one statute, two weeks — the replication at scale has not been run and would settle a live constitutional argument cheaply.

Speculative Is forking secession, exit, or amendment? The 2016 partition produced two chains with a shared history and incompatible constitutional commitments, and no theory in the literature says which it was. If forking is exit, every protocol constitution is voluntary in a sense no state constitution is, and entrenchment is unavailable in principle. Speculative If it is secession, the question is when a minority can carry enough of the polity to survive — and the one observation, Ethereum Classic, then lost its defining property to purchased hashpower.

Frontier Can rights be entrenched against an automated system by procedure alone? GDPR Article 22 has been in force since May 2018 and got its first CJEU ruling on 7 December 2023, in SCHUFA Holding (Scoring), C-634/21 — five and a half years later — holding a credit score is itself a “decision” under Article 22(1) where a third party draws strongly on it. Frontier Commentary in the Industrial Law Journal credits it with closing an accountability gap across corporate boundaries while finding it leaves unresolved whether Article 22 protects against potential or only actual harm, notes the contractual-necessity exception in Article 22(2)(a) severely limits practical workplace application, and calls “similarly significant” inadequately defined. Established No aggregate figures on Article 22 enforcement, invocations or remedies were located in this pass, and that absence is the finding rather than a gap to fill with the court's docket.

Speculative And the minority position that deserves stating because it may be right: the framing is a category error. The useful question may be the reverse one — not whether code can carry constitutional function, but what function must be reserved to institutions that can be argued with: a court that can hear a case the drafters did not anticipate, a legislature that can act on a consequence nobody encoded. Handwave On this reading the correct architecture is code executing content under an analogue wrapper holding the amendment power, which is what every functioning system in this brief already is; the field's advocacy is then a proposal to remove the part that works.

Established The binding bottleneck is not technical: the people who deploy an upgrade, hold the multisig or write the client are outside the system they enforce. Code binds users superbly and rule-changers not at all. Every failure in section 2 has this shape — developers forking a chain with no amendment authority, a foundation moving treasury before its own ratification vote closed, an accumulator buying a supermajority on the open market. That is precisely the position a constitution exists to abolish, and no deployed system abolishes it.

Established The second bottleneck is interpretive determinacy, and it is why no jurisdiction has enacted machine-readable rules as authoritative law. The New Zealand Law Foundation report by Barraclough, Fraser and Barnes finds that legislation-as-code leaving “no room for interpretation at all” “should ring significant constitutional alarm bells”; that “only the Judiciary makes authoritative interpretations”; that existing work shows “little attention to the role of the separation of powers, the independence of Parliamentary Counsel, case law”; and that a coded model must distinguish rules drawn from law itself from rules drawn from non-legal sources — a live risk, because operational policy interpreting a statute acquires equal weight to the statute once both sit inside the same system. Established Their recommendation is decision support, not authoritative legal determinations. The absence of authoritative coded law is a decision taken repeatedly on constitutional grounds, not an engineering backlog.

Established Third: the franchise is a by-product of a technical convenience. Tezos restricts the vote to bakers, so a holder who delegates for block production has delegated their constitutional voice as a side effect — the report commissioned by the ecosystem's own principal developer says so in terms, and Zukowski's 13 of 18 protocols amplifying concentration by 2.45× to 25.65× is the same mechanism measured across the field.

Frontier Fourth: legal personality, which the courts have settled against the code. In Sarcuni v bZx DAO (S.D. Cal., 22-cv-0618, 27 March 2023), arising from a $55 million loss after a phishing attack on a developer's keys, the court declined to dismiss and treated DAO members as a general partnership, reasoning from the fact that token holders “could vote on governance proposals, including proposing to spend treasury funds”. Established Participating in the on-chain constitution was the evidence used to strip participants of limited liability, and the CFTC obtained a default judgment against Ooki DAO in 2023 on a related theory. Speculative Wyoming's 2021 DAO LLC supplement was written to solve this; no authoritative registration count was located from the Secretary of State and none is asserted here, and the best-known critique of the statute predates its commencement, making it a prediction of low uptake rather than an observation.

Established Fifth, and least discussed: nothing in these systems places a substantive limit on what an amendment may do. Tezos's five periods regulate how the protocol changes and say nothing about what may not be changed; Compound's threshold regulates how much delegated weight moves treasury and nothing about whether treasury may move to a proposer's own vehicle. A procedure with no entrenched content is an amendment mechanism, not a constitution — and every deployed on-chain governance system in this record is of that kind.

Established Nothing on this map produces a result this brief waits on. The evidence base is already richer than most of this category, and what is missing is not a discovery. Three constraints are recorded as typed requirements in section 13 and all three are about enforceability: a mechanism binds only if it binds the people able to rewrite the mechanism. Established None is a research question; each is something a legislature, a court or a protocol's founders could choose to supply and has not. Established What the brief does wait on from research is narrow: a replication of the inter-coder agreement experiment at a scale that would tell a drafting office whether an executable encoding is reproducible across independent teams; an independent evaluation of any national Rules as Code programme against outcomes for claimants; and an aggregate of GDPR Article 22 enforcement, which does not appear to exist in any published form — a right in force for eight years with no published measure of its use.

Established The Rules as Code evaluation literature is thinner than its policy footprint by an order of magnitude, and it is worth being blunt about how thin. This pass found no independent evaluation — no audit office, no statutory reviewer, no controlled study — of any national programme's outcomes for citizens. What exists is implementation reports by implementers, an OECD paper by the field's chief advocates, one critical law-reform report, and one small inter-coder study; France's usage figures are self-reported by the programme's own government lab. Speculative Nobody has published a defensible estimate of whether Rules as Code improved decision accuracy, reduced error rates, or changed outcomes for any claimant population. A brief asserting that it works, or that it fails, would go beyond the evidence in either direction.

Established The cheapest decisive experiment is the inter-coder replication, and it would settle a constitutional argument for the price of a research assistant. Witt and colleagues showed divergent interpretive choices surviving an agreed vocabulary over two weeks on one statute. Run the same design with independent teams across statutes of different drafting styles and measure the residual divergence. If a coded encoding is not reproducible across competent teams, the authoritative-coded-law proposal is finished on the evidence rather than the argument; if it is, the strongest objection in the New Zealand report loses its factual basis.

Frontier Second: enact machine-readable rules as authoritative in one narrow area with an explicit judicial-review carve-out, and evaluate it. No jurisdiction has done this, so the question of whether authoritative code degrades legality is entirely theoretical. A rates rebate, an annual-leave calculation, a single benefit taper — all coded already, without anyone taking the last step. Established New Zealand's three-week Better Rules sprint in early 2018 produced decision models and Python for parts of the Rates Rebate Act 1973 and the annual-leave provisions of the Holidays Act 2003, stating the Acts were used “as instruments of exploration”, that “not all legislation is suitable for machine consumption”, and leaving the authoritative-source question open rather than answering it.

Frontier Third: publish the counterfactual for an operating encoding. France's OpenFisca encodes socio-fiscal rules, powers the Assemblée nationale's LexImpact simulator, employment-service estimators and municipal services in four cities, and has been adopted in more than a dozen jurisdictions; its government lab reports about 700,000 simulations for youth assistance in 2022, 122 parliamentary simulations used in debates in 2021 and 1.6 million households informed of entitlements through MesAides in 2019. Frontier Those are self-reported usage counts with no error rate, no accuracy audit and no comparison against the caseworker decisions they displaced. One audit of coded outputs against adjudicated entitlements would convert the largest deployment in the field from an anecdote into evidence.

Frontier Fourth: run the entrenchment test on a live protocol. Write a clause a protocol's own developers cannot amend without a threshold they do not control, then attempt the amendment. Speculative A cheaper observational version exists: publish the identity and concentration of the actors who actually shipped each of Tezos's 21 amendments and compare that set to the set that voted.

Established Fifth: the Canadian regime is already a running experiment and nobody has evaluated its effect on decisions. Its Algorithmic Impact Assessment is mandatory, four-tiered from “little to no impact” to “very high impact”, and published to the Open Government Portal, where 39 assessments have appeared as of August 2026 over about seven years — the largest contributors Immigration, Refugees and Citizenship Canada with 13 and Employment and Social Development Canada with 9. The outstanding measurement is whether filing an assessment changed a system's design or a claimant's outcome, and the register makes that comparison possible for the first time.

Established The one amendment procedure that works mechanically is worth reading as a specification, because its parameters are public and its outturns are counted. Tezos runs five sequential periods of 14 baking cycles, roughly 14 days each: Proposal, Exploration, Cooldown, Promotion, Adoption. A proposal must clear 5% of participating weight; Exploration and Promotion each require an 80% supermajority of Yea over Yea-plus-Nay; and the quorum floats on an exponential moving average of past participation, bounded between 20% and 70%. Established The floating quorum is the most interesting piece of constitutional engineering in the corpus: a fixed quorum in a low-participation electorate either blocks everything or means nothing, and an EMA-tracked one adapts to the polity it actually has. Frontier It also has an unremarked property — a quorum that adapts to falling participation legitimates falling participation — and nothing in the record models what it converges to. Established The thresholds bite: Brest A drew 0.35% and 0.26%; Carthage's first submission reached 3.5% against the 5% floor and failed. Frontier But a proposal failing at 3.5% and one passing are separated by movements well inside the top three bakers' combined 25%.

Frontier The most elaborate purpose-designed digital constitution is Optimism's, and its own operators publish the negative results. The Collective is bicameral — a Token House and a Citizens' House — a separation-of-powers design rather than a voting-mechanism choice. Its governance forum states that “we don't (yet) have the data to show that retroactive funding produces superior outcomes”, that mega-rounds “devolve into popularity contests”, that Round 3 voters “felt overwhelmed”, and that Round 5 created “a perverse incentive to divide work across multiple smaller contributions/teams”. Established It gives project counts and no participation figures at all. Weighted heavily because the interest runs against the finding.

Frontier The dispute-resolution layer is the piece most often assumed and least often built. Kleros, the surviving decentralised court, had resolved about 1,000 disputes with 14,000 registered users as of November 2021, with its own governance token concentrated at a top holder of 11.6%; Zhuk concludes it cannot replace courts, because jurors' qualifications are unverified, appeals are effectively unlimited, and access requirements make it “discriminatory”. Established The more ambitious attempt failed outright: Aragon set out to build an opt-in court and legal order for DAOs and on 2 November 2023 announced its own dissolution, deploying 86,343 ETH — about 87% of non-native treasury — for token redemption with $11 million reserved. Its own post-mortem concedes that “failed attempts at modifying the governance increased tensions” and that “the shaky foundations underlying the current structure cannot be fixed”, naming the court and appchain initiatives among the failures.

Established On the state side, the engineering question is not the platform but the amendment channel the output is routed into — and three cases settle it, in inverse order to their participation volume. Iceland's Constitutional Council received 350 submissions from 218 individuals and associations and, per Popescu and Loveland, 311 discussion threads generating roughly 1,479 comments; a wider figure of about 3,600 comments across all channels is Gylfason's, and he sat on the Council. Frontier Their deliberative measurements need their scope stated: 97% of posts responded to seed posts across the three threads they analysed, while 47% of threads containing two or fewer posts holds corpus-wide. They conclude the exercise was “more accurately described as an exercise in open democracy and transparency rather than crowdsourcing”, with expert lawyers reshaping provisions “beyond mere wording”. Established The referendum of 20 October 2012 returned 48.7% turnout with 66.9% for the draft as the basis of a new constitution and majorities from 66.5% to 82.9% on the five substantive questions. All six passed and nothing was enacted. What killed it was the pre-existing amendment rule — the 1944 constitution requires passage by two successive parliaments with an election in between — and, on Gylfason's participant account, a proposed ratification bar of two-thirds of parliament plus 40% of the electorate, implying roughly 80% turnout to pass anything.

Established Chile ran the largest digital constitutional participation exercise on record and lost twice. The Convention's iniciativa popular de norma platform published 2,456 proposals, drew 2,809,751 signatures from 980,332 distinct individuals and over 2.7 million accesses, and produced 78 initiatives clearing the 15,000-signature threshold by the close on 1 February 2022. The text was rejected 62% to 38% on 4 September 2022; the successor, drafted by a council with a right-wing three-fifths majority, was rejected 55.76% to 44.24% on 17 December 2023 at 84.48% turnout. Two processes, opposite ideological directions, both rejected. Established Brazil's Marco Civil consultation ran two phases — 686 comments in late 2009 and 1,141 contributions in 2010, roughly 1,800 textual contributions. Compared like for like that is 1,800 Brazilian contributions against 2,456 Chilean published proposals — the same order of magnitude; the Chilean signature count of 2.8 million is a different unit and should not be set against it. More than half of first-phase contributions were incorporated, almost half of evaluated second-phase proposals reached the bill, and it became Law 12.965/2014. Established The determinant is not the platform, the volume or the deliberative quality: Iceland's output went into an unachievable amendment channel, Chile's into a plebiscite it lost twice, and Brazil's into ordinary legislation, where it passed.

Established And constitutional courts have not uniformly obstructed digital instruments, which is worth recording against the run of this brief. The Estonian Supreme Court, in constitutional judgment 3-4-1-13-05 of 1 September 2005, dismissed a presidential challenge to unlimited electronic re-voting, holding that uniformity means “all voters must have equal possibilities to influence the voting results” rather than identical procedure, and that re-voting is protective of secrecy and freedom. An entrenched election principle was tested against an electronic procedure and the procedure won on the merits.

The seam that matters most runs against Distributed Governance, and it needs stating in both directions because both slots point at the same artefacts. Distributed Governance owns how a group decides: preference aggregation, token-weighted voting, quadratic voting, futarchy, conviction voting, delegation and delegate markets, DAO treasury allocation, and whether one-token-one-vote produces plutocracy relative to one-person-one-vote. Digital constitutional systems own the higher-order rules constraining what decisions are legitimate at all: amendment procedures and their thresholds, what is placed beyond amendment and by what mechanism, rights entrenchment and its enforceability, judicial review of coded decisions, separation of the powers to write, execute and interpret code, and the legal authority of a coded rule against its natural-language original. The rules for changing the rules.

On the shared objects: Tezos turnout and baker concentration is an electorate-quality question and belongs there; the five-period procedure, the 80% supermajority, the floating quorum and the absence of any substantive limit on what an amendment may do belong here. The carbonvote as a voting mechanism is theirs; the 2016 fork as a constitutional crisis resolved outside every stated procedure, producing an irreversible partition of a polity, is this slot's central case. Compound's Proposal 289 is theirs as “token-weighted voting is capturable by open-market accumulation” and ours as “the amendment rule performed correctly and the outcome was still illegitimate”. Optimism's Citizens' House as an aggregation experiment is theirs; its bicameralism as a separation-of-powers design is ours.

What this slot additionally owns, and Distributed Governance does not touch: Rules as Code and machine-readable legislation; judicial review of algorithmic administration; constitutional courts ruling on digital instruments; state constitution-making with digital participation; GDPR Article 22 and AI Act entrenchment; smart-contract legal effect; and DAO legal personality and member liability. Digital identity, credentials and e-residency are left to the digital-citizenship slot being authored alongside this one; Estonia appears here only as a constitutional court testing an electronic procedure against entrenched election principles, not as an identity system.

Elsewhere on this map: Future Legal Systems, which owns adjudication as such where this brief owns only judicial review of coded decisions; AI Governance and AI-Assisted Governance, which own the systems the AI Act regulates where this brief owns the entrenchment attempt; Institutional Design, where the bind-the-rule-changer problem is stated generally; Future Public Administration, whose evaluation findings explain why no Rules as Code programme has been audited; Future Democracies and Civic Technology, which own the referendum machinery and the participation platforms; and Long-Term Institutions, which asks how any rule survives its drafters. Outside it: constitutional theory and the literature on entrenchment; administrative law; the law of electronic transactions; and mechanism design.

Established The largest entrenchment attempt in force anywhere is not on a blockchain, and it has slipped. The EU AI Act's obligations for stand-alone high-risk systems under Annex III move from 2 August 2026 to 2 December 2027, and embedded systems under Annex I from 2 August 2027 to 2 August 2028, under a digital omnibus that reached political agreement on 6 May 2026 with Member State confirmation on 13 May 2026; AI-literacy duties were softened from guaranteeing a level of literacy to supporting its development. Established There is currently no measured outcome from the AI Act's automated-decision provisions, because they are not yet in force — and the deadlines should be re-checked against the Official Journal before anyone relies on them.

Established The longest-running binding instrument of this kind is Canada's, and its most useful findings are the administering department's own findings against it. The third review, concluded April 2023, found the Directive on Automated Decision-Making did not cover internal services or systems assessing administrative decisions, that six-month review cycles were unworkable, that bias-testing requirements were insufficient, and that it “lacked clear criteria supporting explanations to affected individuals”. The fourth review, in 2024–25, proposes extending it to Agents of Parliament, who are currently excluded. Established Thirty-nine published impact assessments in seven years is a real disclosure regime and a small one, and the department saying so about its own instrument is the strongest evidence in the section.

Established Several load-bearing sources are interested parties and the direction of the interest changes their weight. Tezos documentation is the project's own; the concentration report was hosted by the ecosystem's principal development company and concedes against its sponsor; Aragon's dissolution notice and Optimism's retrospective are operators conceding failure about their own mechanisms; the OECD's Cracking the Code highlights document, from the field's chief institutional promoter, concedes that “most existing RaC initiatives have been experimental and have yet to be scaled”, that implementation “remains strongly contested”, and that RaC “does not promise a panacea”; France's usage figures come from the government lab promoting the programme; Chainalysis sells analytics into the market it measures; and Gylfason's Icelandic account is a participant's. Established The findings that survive their authors' interests are weighted hardest, and in this subject most of them run against the framing.

Established The institutional requirement that would change the picture is jurisdiction over the amender. Every effective remedy in this brief reached a person or an agency: a State ordered to stop using SyRI, a Commonwealth conceding it had no legal basis for $1.763 billion of debts, a tax administration fined €6.45 million across two decisions. Frontier No mechanism in the record reaches a client developer, a multisig signer or a foundation as such, except by treating the whole membership as a general partnership — which reaches the wrong people, punishes participation, and is the outcome Sarcuni actually produced.

Established The harms in this record fall on people least able to contest them, and they are documented rather than inferred. SyRI was deployed in poorer, higher-migrant-population neighbourhoods — Capelle aan den IJssel, Eindhoven, Schalkwijk, Hillesluis and Bloemhof. The Dutch tax administration used nationality as a risk indicator against a population containing about 1.4 million registered dual nationals, and ran a blacklist of 270,000 people with no statutory basis and application by “nationality and physical appearance”. Robodebt's settlement covered approximately 648,000 group members. Established In every case the system's opacity was part of the harm, and in the Dutch case the refusal to disclose is exactly what the court condemned.

Established That has an evidential consequence people get wrong in both directions. SyRI was struck down for opacity and disproportionality under Article 8 ECHR, not for inaccuracy, and no figures were located for how many SyRI projects ran or what they produced. The effectiveness question is unanswered rather than answered negatively, and the reason it is unanswered is the same conduct the court found unlawful.

Frontier Decentralised justice reproduces the access problem it was built to solve. Kleros requires stake to serve, does not verify juror qualifications, and permits effectively unlimited appeals. Established A court whose jurors are selected by capacity to post a bond has an entrance qualification of a kind abolished in most jurisdictions in the nineteenth century.

Established The participation cases raise a distinct question about consent that the enthusiasm around them skips. Iceland ran a genuine public process, passed all six referendum questions and enacted nothing — and the people who blocked it were operating a rule that predated the exercise and was entirely legitimate. Chile put two texts to two referendums, at 84.48% turnout for the second, and the public rejected both. Frontier The ethical case is not that participation was ignored but that it was invited into a channel whose exit was controlled by someone else, which is a specific and repeatable way of spending public trust.

Frontier And on the code side, the exposure runs to participants rather than operators. Sarcuni converted governance participation into evidence of partnership, so a token holder who voted may carry unlimited joint liability for a $55 million loss caused by a phishing attack on someone else's keys. Established A constitution that increases the personal risk of exercising the franchise it grants is inverted with respect to what a constitution is for.

Established The terminal position is that the framing is wrong in an interesting way rather than simply wrong, and the distinction is between constitutional content and constitutional function. Code carries content well, unambiguously: Tezos really has amended itself 21 times in seven years without a fork; OpenFisca really does execute French benefit law across a million-plus households; Canada's regime really has forced 39 disclosures out of federal departments. Established What code does not yet carry is function, because function requires binding the amender, and in every measured instance the amender was unbound. Digital constitutional systems are, so far, legislation with unusually good execution.

Frontier The one dissent from that verdict is SegWit2x, and this brief does not suppress it. A change agreed by actors holding the overwhelming majority of hashing power was abandoned on 8 November 2017 because those with no formal power would not run it. Frontier The correct reading is that the constraint was real, was constitutional in effect, and was not encoded. It came from a distributed user base's ability to exit — to keep running the old rules and take the name and the market with them. That check is unavailable in any permissioned or stake-weighted system, and available in Bitcoin only because Bitcoin never specified an amendment procedure to override it. The lesson a designer should take is uncomfortable: the strongest observed check on rule-changers came from the absence of a rule, not the presence of one.

Established The long-run implication is about what forking does to a polity. The 2016 partition was permanent: two chains, one history, incompatible commitments, and one of them subsequently lost its defining property to purchased hashpower. A constitutional order whose disagreement procedure is duplication rather than decision cannot accumulate what constitutions accumulate, which is a settled record of what was decided and why. Speculative A civilisation running its institutions this way would have no constitutional history, only a set of surviving branches.

Frontier The stable architecture visible in the record is hybrid, and nobody advocates for it because it belongs to nobody's programme: coded execution under an analogue amendment power, with a court retaining jurisdiction over the executors. That is what OpenFisca is, what the Canadian directive is, and what the New Zealand law-reform report recommends. Established No jurisdiction anywhere has enacted machine-readable rules as the authoritative version of its law, after roughly eight years of well-funded programmes in New Zealand, France, Australia and Canada. That is not an engineering backlog; it is a decision taken repeatedly by the constitutional lawyers asked to look at it.

Speculative And the strongest speculative version of the field's own claim deserves stating, because if anything rescues the framing it is this. A constitution binds rule-changers through anticipated cost, not through physics. If a coded entrenchment clause were embedded in a system whose value depended on it — so that overriding it destroyed the thing worth capturing — the binding would be economic rather than legal, and would in principle reach the developers. Handwave Ethereum Classic is the one attempt at exactly that and it failed at a cost of about two days of mining, with the price moving under 4%. The mechanism is coherent; the single test went the wrong way; and the step from “economically costly to override” to “therefore binding” is where this argument does its work by assertion.

These horizons track statutory commencement dates, protocol amendment cycles and litigation rather than technology:

• 10 yr: Established Three dates are hard: the EU AI Act's stand-alone high-risk obligations apply from 2 December 2027 and embedded ones from 2 August 2028, so the first measured outcome from the largest entrenchment attempt in the world arrives late in this window and not before; Canada's fourth review lands with a decision on whether Agents of Parliament come into scope; and Tezos either continues amending on its published schedule or stops, either of which is informative. Frontier Expect at least one further court to treat DAO governance participation as evidence of partnership, because Sarcuni's reasoning transfers directly and no legislature has displaced it outside Wyoming. Frontier Expect no jurisdiction to enact machine-readable rules as authoritative law in this window — the constraint is the separation of powers and it has not moved in eight years. Speculative The live hypothesis worth watching is whether an on-chain entrenchment clause is tested against a core development team in earnest; nothing predicts when, and the first such case would be the most informative event this subject could have.
• 25 yr: Speculative If the hybrid architecture is the stable one, the trajectory is coded execution spreading fast and coded authority not spreading at all, with judicial review of algorithmic administration becoming an ordinary branch of administrative law — the four judgments in section 2 are already a doctrine in embryo. Speculative The competing hypothesis is that a jurisdiction takes the last step in one narrow area, a benefit taper or a rebate calculation, and the coded version becomes authoritative with judicial review preserved; Frontier that is a live proposal with advocates and no precedent, and the inter-coder reproducibility result would decide it. Speculative A third possibility, less discussed and not refuted: protocol constitutions converge on legal wrappers, the on-chain layer becomes execution only, and the constitutional question migrates entirely back to the courts that already answer it.
• 50 yr: Speculative At this horizon the question is whether any encoded entrenchment has ever bound an amender, and the record contains no instance and one near-miss produced by the absence of a procedure. Speculative If the exit mechanism that stopped SegWit2x is the durable one, the design implication reverses the field's programme: leave the amendment procedure unspecified and preserve cheap forking, which maximises the veto and destroys any possibility of a settled constitutional record. Handwave Choosing between those is asserting a social outcome, not extrapolating a measurement.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The instruments that did the constraining here date from 1950, 1947 and 1789 as read through the bloc de constitutionnalité, and the oldest artefact on the coded side is seventeen years old. Handwave A subject whose analogue instruments outlive its digital ones by two centuries has no base rate at this horizon, and the only defensible statement is that the durability comparison has not begun.

• Depends on Nothing on this map. This brief waits on no result another brief produces: the constitutional record is already better documented than most of this category, and what is missing is enforceability rather than knowledge. No typed depends-on edge is claimed.
• Requires (not on this map) All three constraints are about enforceability, and each names a mechanism that would have to bind the people able to rewrite the mechanism. First, an amendment rule reaching the client developers who ship the upgrade: the 2016 Ethereum fork was executed by developers holding no formal amendment authority, four days after a non-binding poll drawing 5.5% of supply in which about a quarter of the affirmative weight came from one address, and no deployed system since has placed those developers inside the procedure they execute. Second, a coded rule a court will treat as authoritative against its natural-language statute: no jurisdiction has taken that step after roughly eight years of programmes in New Zealand, France, Australia and Canada, because the law-reform position that “only the Judiciary makes authoritative interpretations” has not been answered, and the one experimental study of encoding found divergent interpretive choices persisting even after coders agreed the key legal terms in advance. Third, a treasury that cannot move before the vote authorising it closes: Arbitrum's Foundation transferred and began spending 750 million ARB while its own ratification vote was heading to defeat, and Compound's Proposal 289 moved 499,000 COMP by 682,191 to 633,636 into a vehicle the proposers controlled. None of the three is a research result; each is something a protocol's founders, a legislature or a court could choose to supply, and none has.
• Enables In principle any coded administration this map assumes — algorithmic benefits assessment, automated regulatory compliance, protocol-governed infrastructure — rests on a constitutional layer that can reach the people who write it. No typed enabling edge is claimed, and the reason is a finding: no such layer has been demonstrated, so nothing measurable is being enabled.
• Adjacent Constitutional theory and the literature on entrenchment; administrative law, which supplies every effective remedy in the record; the law of electronic transactions, which supplies smart-contract legal effect; mechanism design; and within this map Distributed Governance, Future Legal Systems, Institutional Design and AI Governance.

Established “Code is law.” The claim was tested in 2016 and failed within a month of the test starting. The DAO's code executed exactly as written, the community declared the outcome illegitimate, and a chain-wide state change was pushed by client developers with no amendment authority four days after a non-binding poll drawing 5.5% of ETH supply. Established Code executed; the constitution was made off chain by people the code did not reach. Every subsequent case in this brief has the same shape.

Established “Blockchains are immutable, and that is what makes them constitutional.” Ethereum Classic exists precisely to entrench immutability as its single commitment, and in August 2020 it was 51%-attacked three times in one month, the largest reorganisation running to more than 7,000 blocks, about two days of mining. Established The chain whose constitution is the impossibility of rewriting history had its history rewritten three times in four weeks, by people who bought hashpower on the open market, and the price moved less than 4%. Frontier A property purchasable at market rates is a cost, not an entrenchment.

Established “Arizona legislated that blockchains are immutable, so the law recognises the property.” What Arizona wrote at §44-7061(E)(1) is a definitional recital attached to an operative provision that does very little, and the correction runs against this brief's own argument: Svikhart concludes the provisions survive federal ESIGN preemption “for now”, because they accord blockchain records equal rather than greater status. The statute is not in trouble. It is merely saying something false in a recital.

Frontier “DAOs are decentralised governance.” Measured concentration says the franchise is narrow and the amendment thresholds narrower: less than 1% of holders control 90% of voting power across the ten DAOs Chainalysis examined, and between one in ten thousand and one in thirty thousand holders can unilaterally carry a proposal. Established Tezos's top three bakers hold about 25% of voting weight and in one 99.61% result the votes of the 71 largest stakeholders were individually deducible; Kleros's top token holder is at 11.6%; Compound's Proposal 289 and Arbitrum's AIP-1 closed the argument in practice.

Established “And therefore DAOs are uniformly plutocratic.” That does not follow, and the best measurement says the popular figures are inflated. Zukowski's 2026 audit of 52 protocols finds top-1,000 concentration spanning HHI 0.005 to 0.199, median around 0.04 once protocol-controlled addresses are stripped out, and prior studies inflated by a median factor of 2.3× and up to about 18×. Polkadot at 0.005 is less concentrated than many national equity markets, and launch allocation predicts nothing about eventual governance power (r = 0.09, p = 0.55, N = 50). Established The robust adverse finding is narrower and different: delegation amplifies concentration in 13 of 18 protocols by 2.45× to 25.65×. Anyone repeating “1% control 90%” as a general truth about DAOs is repeating a 2022 vendor blog about ten hand-picked tokens.

Frontier “SegWit2x proves code can govern.” It proves something, and not what it is usually cited for. An upgrade agreed by parties holding the overwhelming majority of hashing power was cancelled on 8 November 2017 for lack of consensus, defeated by node operators and users with no formal lever at all. Established That is a genuine constitutional outcome and it was not produced by any rule: Bitcoin has no amendment procedure, nothing counted the refusal, and what operated was a convention enforced by the ability to exit and take the name. Frontier It is also one-directional — it can block a change, never compel one, and nothing constrains the blockers. It belongs in the record as the strongest counter-case and it does not restore the framing.

Established “Estonia has a robot judge.” It does not, and this is among the most-cited claims in the field. The Ministry of Justice stated on 16 February 2022 that Estonia “does not develop AI robot judge for small claims procedure nor general court procedures to replace the human judge”, calling the March 2019 reporting misleading. Established What Estonia is doing is automating the order-for-payment procedure, roughly 50% of civil cases, plus transcription and anonymisation. A government correcting reporting that flattered it is about as reliable as a denial gets.

Established “The EU AI Act shows rights can be entrenched against automated decisions.” Nothing has been shown, because nothing is in force: high-risk obligations moved to 2 December 2027 and 2 August 2028 under an omnibus that reached political agreement on 6 May 2026 with Member State confirmation on 13 May 2026. Established Any claim of a measured AI Act outcome today is a claim about an instrument that has not applied to anyone. Frontier The comparable claim for GDPR Article 22 is weaker than usually stated for a different reason: it produced its first CJEU ruling five and a half years after application, and no aggregate of enforcement actions or remedies was located in this pass.

Speculative “Rules as Code works” and “Rules as Code has failed” are both beyond the evidence. No independent evaluation of any national programme's outcomes for citizens was located — no audit office, no statutory reviewer, no controlled study. What exists is implementation reports by implementers, an OECD paper by the field's promoters that itself concedes the initiatives remain “experimental and have yet to be scaled”, one critical law-reform report, and one small inter-coder study. Established Nobody has published a defensible estimate of whether Rules as Code improved decision accuracy, reduced error rates, or changed outcomes for any claimant population.

Established Three further figures are deliberately not printed, and naming them stops their absence reading as an oversight. No authoritative Wyoming DAO LLC registration count was located from the Secretary of State, and the best-known critique of that statute predates its commencement, making it a prediction of low uptake rather than an observation. Established SyRI's effectiveness is unmeasured, not disproved — the court struck it down for opacity and disproportionality, not inaccuracy, and no figures exist for how many projects ran, because non-disclosure is what the court condemned. Frontier And Gylfason's Icelandic account, including the roughly 3,600-comment total that circulates as though it were an independent measurement, is a participant's; the structural facts around it hold independently.

Established And the framing itself: a constitution's distinguishing function is not to state rules but to bind the rule-changers. Statutes state rules. Code executes them, and executes them better than any statute. Established Across the whole measured record — four judicial interventions that stopped automated administration, one user revolt that stopped a protocol upgrade, and zero cases of an encoded rule constraining the people able to change it — the framing survives as a claim about content and fails as a claim about function. Code can carry constitutional content. It has not yet carried constitutional function, because the people who write, deploy and upgrade the code have in every measured instance stood outside the system they were enforcing.