Distributed Governance

Distributed governance means making binding collective decisions without a central authority, in organisations whose members join rather than reside. That covers five families with very little in common except the absence of a centre: token-weighted decentralised organisations, the commons institutions Elinor Ostrom's programme documented, worker cooperatives and mutuals, open-source projects, and the protocol and standards bodies — the IETF, ICANN, and now the federated social networks — that govern shared infrastructure nobody owns.

The subject has an unusual evidentiary property that makes it worth taking seriously: much of it is measurable in a way conventional governance is not. On-chain voting is a public record, so turnout, concentration, delegate power and pass rates can be computed exactly rather than surveyed. Git histories are a public record, so the real distribution of authorship in an open-source project can be computed rather than asserted. Wiki revision logs are complete populations, not samples. This is the only area of political science where the underlying data are usually total rather than sampled, and that is the reason this brief can state numbers where the neighbouring briefs state findings.

Where this brief stops. Future Federalism owns territorial multi-level government: equalisation formulas, fiscal decentralisation, devolution settlements, subnational debt, secession, and the tier structure of states. This brief owns governance that is non-territorial and non-hierarchical — membership-based rather than residence-based, voluntary rather than compulsory. The arbitration rule where they meet is whether the governed unit is a jurisdiction or an association. Ostrom's eighth design principle, nested enterprises, straddles the line: where the nesting runs between tiers of government it belongs next door, and where it runs between user associations, irrigator councils or cooperative federations it belongs here. Mondragón is here. Catalonia is there.

The subject is also heavily colonised by cryptocurrency advocacy, and this brief's contribution is the opposite of advocacy. Nearly every figure below comes either from academics with no protocol affiliation computing from public chains, from complete-population software-repository studies, from a securities regulator's investigation, from federal court judgments, or from institutions writing critically about themselves. Where a source has an interest in the outcome — a sector federation reporting cooperative survival rates, a crypto-data firm reporting a token burn, a standards body assessing its own effectiveness — the reading list says so and says what is and is not relied on.

The one-sentence result: the forms of distributed governance that have worked for decades all have a membership boundary that money cannot cross, and the form with the worst measured concentration is the one whose only qualification for membership is purchase. That is Ostrom's first design principle, restated as an empirical regularity across five very different institutional families, and it is what the rest of this brief is about.

Established Concentration in token-weighted governance is extreme in the flagship cases and bimodal across the population, and reporting either figure alone is wrong. In the five most-studied protocols on data to 19 June 2025, token-holding Gini coefficients run above 0.99 for Aave, Compound, ENS and Uniswap, and 0.75 for Nouns, whose tokens are distributed by daily auction rather than airdrop. Voting-power Gini exceeds 0.94 everywhere and reaches 0.99 for Compound, ENS and Uniswap. Nakamoto coefficients at the 33 to 50% control threshold: ENS 1 to 8, Uniswap 8, Compound 19; among delegates at the 50% threshold, Uniswap 16 and Compound 13. For ENS a single delegate controlled between a third and a half of all delegated power. But across a sample of 100 organisations formed between 2019 and 2023, drawn from an ecosystem of more than 13,000 holding $24.5 billion of treasury among 11.1 million token holders, the mean holder Gini is 0.794 with a standard deviation of 0.275 and the mean Nakamoto coefficient is 35.9 with a standard deviation of 120.4 across a range of 1 to 1,046. A standard deviation more than three times the mean is not a concentrated population; it is a bimodal one containing a few genuinely broad organisations and a mass of near-degenerate ones. The right statement is that the most-studied protocols are the most concentrated protocols, and that 42% of organisations hold under $1 million, so most of what is called distributed governance is a group chat with a treasury.

Established Turnout is correspondingly thin. Share of token supply voting per proposal runs 3.2% to 7.7%. Average voter counts per proposal: 121.8, 72.0, 773.3 — and, in one large protocol, 6.5. In the largest studied polling record, a mean of 24.6 voters per poll across 638 polls, with the single largest voter holding an average 52.7% of votes cast and poll-level Gini never falling below 0.58. Across the hundred-organisation sample, voter wealth and participation rate are significantly positively correlated in many cases — the rich vote more often, so measured turnout understates measured concentration rather than offsetting it.

Established The finding that inverts the obvious reading, retained from the previous version of this brief because it is still correct. Large delegates voted differently from small delegates on 6 of 84 proposals in one protocol and 1 of 10 in another. The researchers' own summary: “we rarely observe these powerful entities overturning a vote by choosing a different outcome than that of the overall community and less influential voters.” Concentration of power is not the same as exercise of capture, and every popular treatment of this subject conflates them.

Established But the flattering reading is wrong too, and a securities regulator got there first. Nearly all proposals pass by large margins because contested questions are filtered out off-chain in informal temperature checks before reaching a vote; in most votes the top three or four holders together constitute a majority. What is measured with such precision is not a decision procedure but a ratification procedure attached to an off-chain deliberation whose participants are not measured at all. The striking thing is that the United States Securities and Exchange Commission reached the same structural diagnosis in July 2017, in its Section 21(a) report on The DAO, and reached it from a hostile direction. Token holders' voting rights did not amount to managerial control, the Commission found, because holders “could only vote on proposals that had been cleared by the Curators”, because they lacked the information for an informed vote, and because “the pseudonymity and dispersion of the DAO Token holders made it difficult for them to join together to effect change.” One Curator stated publicly that he had “complete control over the whitelist.” An enforcement agency and a sympathetic academic literature converging independently, nine years apart, on the same account of where the decision actually happens is stronger evidence than either alone.

Established The founding case establishes that concentration is an initial condition, not a degeneration. The DAO launched on 30 April 2016 and raised roughly 12 million ETH — about $150 million — selling about 1.15 billion tokens to more than 11,000 investors in a 28-day offering, taking in nearly 14% of all ether then in existence. As of 17 May 2016, three weeks in, the largest single investor held under 4% of tokens and the top 100 holders controlled just over 46%. No delegation, no vote-buying, no whale accumulation over time: a purchase-based membership boundary produced near-majority control by a hundred accounts before anyone had voted on anything.

Established And the recovery from the first great failure required an authority that had no governance role at all. On 17 June 2016 a reentrancy exploit drained 3.6 million ETH, roughly one third of the 11.5 million ETH committed. On 20 July 2016 the Ethereum chain was hard-forked to return the funds, and the minority who refused became Ethereum Classic. The lesson usually drawn is that code is not law. The sharper lesson is that the organisation had no mechanism to reverse the theft, and what reversed it was the client developers and miners of the layer beneath — a body nobody had designed as a governance body and nobody had elected. Every subsequent governance failure has been resolved the same way: by an ad hoc discretionary authority appearing from outside the rules. The flash-loan attack that drained $181 million for roughly $77 million of net profit ended with the protocol removing on-chain governance entirely and replacing it with a multisig wallet. A second attacker passed a proposal containing a hidden function minting himself 1.2 million votes, because the organisation approved code it had not read. And in the case that matters most, no rule was broken at all: a coordinated group passed a valid proposal moving roughly $24 million, about 5% of a treasury, into a yield product they controlled, on a margin of about 48,500 votes, with voting opening late on a Thursday and closing over a weekend. That is the difference between an exploit and a governance outcome you dislike, and the distributed model has no mechanism for the distinction.

Established The legal-personality question has been answered, and answered against the organisations. On 8 June 2023 the United States District Court for the Northern District of California entered default judgment for the Commodity Futures Trading Commission against Ooki DAO: a $643,542 civil monetary penalty and a permanent injunction, on the holding that the organisation was “an entity capable of being sued” as an unincorporated association under California law, because token holders had voting control over protocol updates, fund management and operations — the court's framing being that this was not merely autonomous software but an entity run by people. Service of process was effected through the organisation's own help chat box and web forum. In Samuels v. Lido DAO, in the same district, the court accepted that the organisation could be pleaded as a general partnership, which if it holds makes governance participants jointly and severally liable without limit. In both, the fact that creates the exposure is that you voted.

Established The commons literature is the strongest quantitative evidence in the subject and it says distributed governance is institutionally demanding rather than light-touch. A review of 91 studies covering 77 cases found Ostrom's design principles — clear boundaries, congruence with local conditions, proportionality, collective-choice arrangements, accountable monitoring, graduated sanctions, conflict resolution, rights to organise, nested enterprises — well supported empirically and reformulated them into eleven sub-principles that are explicitly probabilistic and diagnostic. Re-coding 69 cases across forestry, irrigation and fisheries and applying qualitative comparative analysis: no single principle predicts success in isolation, four are necessary but insufficient, sufficiency requires configurations of ten or eleven, and fisheries required all eleven. Token-weighted governance routinely invokes this literature while satisfying perhaps three: no graduated sanctions, no conflict-resolution mechanism, no accountable monitoring, and no meaningful boundary at all, since anyone can buy in.

Established That last point is this brief's central claim, and it holds across five institutional families that share nothing else. The forms of distributed governance with a real multi-decade record all have a membership boundary money cannot cross: you become an appropriator in a commons by being one, a worker-member by working and buying in on regulated terms, a Python core developer by earning commit rights, an IETF participant by turning up and being technically persuasive. Token governance does not fail to satisfy Ostrom's first principle. It is constituted to violate it, because the transferability that makes a governance token a liquid asset is exactly what makes the membership boundary purchasable. The concentration figures are the arithmetic consequence rather than an implementation failure.

Established The cleanest evidence for that claim is a turnout comparison nobody makes. When Python's core team elected the Steering Council that replaced a founder who had governed for twenty-seven years, 69 of 96 eligible voters took part — 71.9% — on a bounded, earned, non-transferable, non-purchasable franchise. Token-weighted organisations, where the only qualification is purchase, turn out 3.2% to 7.7% of supply and a mean of 24.6 voters per poll. The comparison is this brief's; both inputs are sourced. An order of magnitude and a half separates a bounded electorate from a purchasable one on the same species of question.

The genuinely open questions in this subject are not the ones the field advertises. Whether a better voting mechanism can be designed is not open in any interesting sense; mechanisms are cheap and the literature is deep. What is open is whether any of these forms can hold authority at scale without reconcentrating, and whether the conditions under which they have worked can be transported to conditions in which they have not been tried.

Established The iron law of oligarchy is a testable empirical claim, it has been tested well, and it survived. This is the most important thing in the brief that was not in the previous version. Shaw and Hill analysed 683 of the largest Wikia wikis with complete revision records from 2004 to April 2010, covering every edit by registered and unregistered users and all administrator activity, and operationalised Michels three ways. A one-log-unit increase in contributors is associated with odds of 0.81 of adding new administrators — growth makes new appointments less likely, not more. A 1% increase in accounts is associated with a 3% rise in administrator edits and a 5% increase in administrator reversions of experienced contributors' edits. Their conclusion: “as wikis become larger, a small group – present at the beginning – monopolizes positions of formal authority in the community and accounts for more administrative activity while also using their authority to restrict contributions from experienced community members.”

Established What makes that result decisive is what the setting lacks. There is no capital in a Wikia wiki, so the mechanism cannot be money. Membership is free and unbounded, so it cannot be an exclusive boundary. The result is on 683 organisations with complete population data, not a case study. The iron law survives the removal of both explanations distributed-governance advocates usually offer for reconcentration. What accumulates is founder tenure and administrative attention, and it accumulates faster than the population grows.

Frontier Whether Ostrom's design principles extend beyond the scale at which they were measured is open, and the person who tried the extension named what breaks. The evidence base was collected on irrigation districts, inshore fisheries and forest user groups of tens to low thousands of members. Extending it to global commons, Stern accepts the framework's external validity and then states that defining boundaries for resources and appropriators “is not a meaningful exercise for global commons” where participants number in the millions; that direct participation by most users is infeasible and the principle must be weakened to meaningful participation by the range of interested and affected parties; that monitoring independence from appropriators becomes essential precisely where it is structurally hardest, because the users fund the monitoring; that local autonomy must yield to higher-level restriction because lower-level actors can externalise costs; and — the sharpest — that learning from direct experience is not a feasible option, because the errors are irreversible. His conclusion is that the principles “require some modification and extension”, not that they fail. That is the honest position. What it rules out is the move both crypto governance and global-commons advocacy make: citing evidence from village-scale institutions in support of governing something with millions of anonymous participants.

Frontier Exit-based governance is a live experiment and its interim result is that the exit exists and is not exercised. The Fediverse and the AT Protocol substitute cheap leaving for effective voice. Measurements from three independent groups agree. On Mastodon, 96% of users join 25% of the largest instances; the November 2022 migration produced 4,908 single-user instances, roughly 40% of them created in that month, most subsequently dormant, and every instance studied except mastodon.social peaked in 2022 and contracted in 2023. On Bluesky, across 5.5 million users and 225 million posts, 98.9% of handles sit under bsky.social, with only 57,202 on alternative domains and half of those registered through four registrars; the relay is operated by Bluesky PBC with no documented competitor; and of 40,398 active feed generators, the top three hosting platforms carry 95.8%, with one carrying 85.86%. Federation is nonetheless real at the protocol layer: mastodon.social users follow accounts across 18,847 distinct instances, and across 713 Pleroma instances 99% of the posts in a conversation are federated rather than local.

Frontier The one component that has genuinely dispersed is moderation, and that asymmetry is the finding. Bluesky had 62 labelling services identified by May 2024, 36 of them actually issuing labels, and by April 2024 community labellers were producing 88.7% of all labels. Moderation is the expensive, unglamorous, legally exposed function. The brief's reading — reasoned inference, not a claim any consulted source makes — is that the components of a federated system that actually decentralise are the ones the incumbent operator has an interest in being rid of. The Pleroma data shows what devolving it means in practice: a single instance produced 23.33% of all toxic posts and the top three about 31%, so the moderation burden is as concentrated as everything else and now falls on volunteers.

Frontier Quadratic voting has a real five-session field deployment and it ended in a court ruling rather than an efficiency finding. A legislative caucus used it to prioritise a backlog of over 100 bills worth $120 million against roughly $40 million of funding, each member receiving 25 tokens at quadratic cost. Participants reported a better signal with less noise, and it was used again across four subsequent sessions. Then a district court held that the anonymous system violated open-meetings law: it constituted a meeting that was not public, functioned as a secret ballot on legislative priorities, and operated as a serial meeting circumventing transparency requirements. The property that makes quadratic voting work — anonymous honest revelation of intensity — is the property that made it unlawful for a legislature. That structural incompatibility appears nowhere in the mechanism-design literature. Handwave And no independent evaluation of whether it produced better budget decisions has ever been published; the only outcome claims come from participants and from the mechanism's own promoters.

Handwave Futarchy remains untested. Twenty-six years after the proposal, the largest real deployment had made single-digit numbers of decisions, and one participant openly committed over $250,000 to move a conditional market in the direction he wanted — the mechanism demonstrably moved by capital rather than informed by it. The one peer-reviewed paper contains no live market data at all: it simulates counterfactuals retrospectively over ten past proposals, reports full concordance, and concedes hindsight bias and selection bias in doing so, with the authors disclosing holdings in the organisations studied. A simulation that assigns outcomes after the fact and then reports 100% agreement is not evidence.

Established The first bottleneck is that a purchasable membership boundary makes concentration an identity rather than a defect. The commons evidence treats clearly defined boundaries as the load-bearing design principle, and token governance's boundary is a market price. That is not a parameter to be tuned: the transferability that makes a governance token a liquid asset is the same property that makes membership purchasable. The DAO's top 100 holders controlled just over 46% within three weeks of launch, before anyone voted on anything. Nothing downstream — delegation, quadratic weighting, quorum design — can undo a boundary condition set at issuance.

Established The second is that the mechanism cannot distinguish an attack from a legitimate outcome it dislikes. That is not a bug to be patched; it follows from the design. A system whose legitimacy rests entirely on rule-following cannot appeal to anything outside the rules when the rules produce a result participants regard as theft. Every response observed so far reintroduces discretion — a multisig, a foundation veto, an off-chain filter, a social consensus to fork, or in the founding case a hard fork of the underlying chain executed by developers and miners with no governance role at all. Which is to say, reintroduces hierarchy at exactly the point where hierarchy was supposed to be unnecessary.

Established The third is that the measurable layer is not the deciding layer. Precise on-chain metrics describe ratification. The actual decisions are made in forums, calls and private channels by participants nobody counts. The Securities and Exchange Commission identified the same structure in 2017 — holders could vote only on proposals the Curators had cleared, and one Curator claimed complete control over the whitelist — and it has not changed. This is the mirror image of conventional governance's problem, and it means the field's evidentiary advantage is narrower than it looks.

Established The fourth is that legal exposure now attaches to participation. An unregistered organisation is an unincorporated association capable of being sued, or pleadable as a general partnership with unlimited joint and several liability, and in both holdings the operative fact is that token holders vote. The statutory cures require registration, a registered agent, a jurisdiction and a public notice — territorial legal anchoring, which is precisely what the form was constituted to escape. There is no configuration in which the organisation has both legal capacity and no jurisdiction.

Established The fifth is that transaction costs exclude the small participant, and the arithmetic is brutal. ConstitutionDAO raised roughly $47 million from 17,437 unique wallets at a median contribution of $206.26. Reported gas costs were about $60 to deposit and an estimated $60 to claim a refund; a separate account cites a contributor who paid $70 to donate $200 and another $70 to withdraw it. The two figures disagree by roughly 15% and this brief does not resolve the discrepancy. On the median contribution, a round trip at $60 each way consumes about 58% of the donation — that arithmetic is this brief's, the inputs are sourced. Weeks after the auction roughly $23 million remained unrefunded, which is the predictable consequence of a refund costing more than a quarter of what it returns. A governance system whose participation fee is a substantial fraction of the median stake has a wealth qualification whether or not it has a rule.

Frontier The sixth is that the largest working example of distributed ownership stays viable by being hierarchical where it counts. The world's largest worker-cooperative federation runs €11.2 billion of sales across more than 70,000 people — and has roughly 30,100 worker-members, meaning fewer than half the people working for it are members of it. Its international plants are largely conventional subsidiaries. And its democratic bodies' most consequential decision in fifty years was to refuse to rescue a founding member cooperative, letting the employer of some 5,700 people fail rather than subsidise it.

Established The seventh, and the one the cooperative literature least expects: the binding constraint at Fagor was not capture but abdication. By 2006 the cooperative employed close to 11,000 people. It had been losing more than €14 million a year in the Spanish refrigerator unit alone since 2005 — fifteen years before the bankruptcy — with further losses in washing machines and dishwashers. The governance diagnosis from inside the Basque academic community is that one-member-one-vote general assemblies were too slow to close loss-making units; that assembly attendance was minimal except when pay cuts were on the agenda; that governing council members “were unable to question the quality of the strategies that managers proposed because they did not have the training, the experience, the knowledge”; and that one-member-one-vote produced a representation bias in which factory workers dominated the councils while marketing and finance were absent. Power that formally belongs to everyone and is exercised by almost nobody flows to whoever is present — here, to management, by default.

Established Nothing here waits on a research result. The cryptography works, the voting contracts work, the commons literature is fifty years old, and the standards bodies have been operating for decades. That is an unusual sentence in a frontier-research corpus and it is the correct one here. What distributed governance waits on is institutional and legal.

Established It waits, first, on a membership boundary that is not simply purchasable — the design principle the strongest evidence base treats as most load-bearing, and the one token governance is constituted to violate. Every other dependency below is downstream of this one, and no mechanism-design work can substitute for it.

Established It waits on graduated sanctions and a conflict-resolution mechanism, the two further design principles the quantitative commons work identifies as necessary and which token governance almost universally lacks; and on accountable monitoring separate from the participants being monitored, which in the federated networks currently means unpaid volunteers absorbing a moderation burden as concentrated as everything else in the system.

Established It waits on a legal wrapper able to hold a treasury and bear liability, which every organisation reaches for the moment something goes wrong, and which is now supplied — at the price of accepting a jurisdiction, a registered agent and a public notice that member rights differ materially from an ordinary company's.

Frontier It waits, weakly, on transaction costs low enough that participation is not wealth-qualified. A round-trip cost of roughly 58% of the median contribution is a franchise restriction implemented in gas rather than in law. Layer-two settlement has reduced this substantially since 2021; no source consulted for this brief measures the current figure, so the improvement is stated as direction rather than magnitude.

Established What depends on it is thinner than the field assumes. No brief in this corpus depends on distributed governance succeeding; the briefs that assume coordination assume it generically rather than through this route. The real dependants are outside the map: any proposal for internet infrastructure governed by no state, any commons-based resource regime, and any plan for a federated alternative to platform monopoly — all of which currently inherit both this brief's evidence and its unresolved questions.

Established The measurement that would settle most of the token question is cheap and nobody publishes it: the off-chain funnel. How many proposals are raised, how many survive the temperature check, who participates in that filter, and what the pass rate looks like before the filter rather than after. Every headline on-chain statistic is conditional on a stage that goes unmeasured, and the regulator identified that stage as decisive in 2017.

Established Open source is the natural experiment already running, and it has returned a negative result about the gap between constitution and production. Across 1,932 popular GitHub projects in six languages, 57% have a truck factor of one — one person whose departure removes knowledge ownership of most of the code. 315 projects, 16%, were abandoned when all core maintainers left, and 66% of those abandonments happened in truck-factor-one projects. Of the 315, 128 (41%) survived because new core developers took over, in 86% of cases a single replacement, and in 64% of cases within the first year. A survey-validated oracle across 35 projects gives the same shape: 57.1% at truck factor one, 71.4% at two or below, with only two systems above ten. An independent replication over the top 1,000 repositories by star count finds nearly half at a bus factor of two or less, only 10% at six or above, and no correlation between popularity and bus factor. The field that supplies distributed governance's best success story has a median project in which one person is the governance.

Established The benevolent-dictator-to-foundation transition is the other natural experiment, and its result is continuity. Python's founder resigned as benevolent dictator in July 2018; the Steering Council model was accepted that December; the first election ran in February 2019 with 96 eligible core developers, 69 of whom voted, choosing five from seventeen nominees under a bloc-voting rule capped at five selections. The result returned the former benevolent dictator and four other long-standing figures, on what the specialist coverage described as complete landslide majorities for the best-known candidates, with the five-vote cap criticised at the time for concentrating support on established figures and the outcome read as choosing the status quo. Formal authority moved and informal authority did not. That is not a criticism of Python, which handled a genuinely hard succession unusually well; it is evidence for the iron law and against the assumption that changing the constitution changes who governs.

Frontier An experiment that would settle the central claim and has not been run: apply the eleven commons design principles as a scored checklist to token-governed organisations and test whether configural satisfaction predicts survival, as it does for irrigation systems and fisheries. The coding instrument exists and has been validated on 69 cases; nobody has pointed it at this domain. The prediction that follows from this brief's position is specific — token organisations should cluster at three or four principles satisfied, and survival should be uncorrelated with anything else measured — and it is falsifiable within a single research project.

Frontier A second: measure whether turnout falls as the general-partnership doctrine propagates. If voting is what makes you liable, rational participants should stop voting, and the on-chain record makes this measurable to the proposal. A clean before-and-after around the Ooki judgment of 8 June 2023 and the Lido ruling is available to anyone with a Snapshot archive.

Frontier A third: an independent evaluation of the quadratic-voting field deployment against the counterfactual. Five sessions of legislative prioritisation is a real dataset and the only outcome claims come from interested parties, one of which publishes no vote totals or rankings at all.

Established A negative result worth recording as an experiment in its own right. ConstitutionDAO is the largest single-purpose distributed-governance event ever run and it failed at every stage after fundraising: it lost an auction to a $43.2 million bid while holding $47 million, because it had to reserve funds to insure, store and transport the object; it had no procedure for deciding what to do with the money once it lost, and dissolved rather than repurpose it; and its refund mechanism consumed a majority of the median contribution. Speculative The brief's own reading of the auction loss — that a crowdfunded bid publishing its balance in real time has published its reserve price, and that radical financial transparency is structurally incompatible with adversarial negotiation — is reasoned extrapolation. No consulted source makes the causal claim, and the organisers' stated reason was the insurance-and-storage reserve.

Established The engineering lessons of on-chain governance are specific and each was learned from a loss. Separate voting from execution with a mandatory timelock, because the flash-loan attack worked by doing both in a single transaction. Require independent review of proposal bytecode rather than its description, because one organisation approved a hidden vote-minting function it had not read. Set quorum and voting windows against adversarial timing rather than convenience, because a treasury proposal that opens late on a Thursday and closes over a weekend has chosen its electorate. And publish delegate concentration continuously, because the pivotal delegator in two major protocols is a single firm and that fact is discoverable only by external analysis.

Frontier One engineering lever has been identified that is neither cryptographic nor procedural: the delegate interface. The ranking-based delegate lists on the dominant governance front-ends reinforce a visibility bias toward delegates who are already powerful — a delegator arriving to delegate sees the largest delegates first and delegates to them, which is a positive feedback loop implemented in a web page. This has no analogue in conventional political science, where ballot-order effects are the nearest thing and are an order of magnitude smaller. It is also the cheapest thing on this list to fix and the least discussed.

Established The legal engineering is now the binding part, and the statutory forms are more revealing than they look. Wyoming's 2021 Decentralized Autonomous Organization Supplement makes a DAO a limited liability company, member-managed or algorithmically managed, requiring “DAO”, “LAO” or “DAO LLC” in the registered name. Membership interest is a member's digital-asset contribution divided by total contributions, and where no digital assets are contributed each member gets one equal vote. Members owe no fiduciary duties except the implied contractual covenant of good faith and fair dealing, and the articles or the smart contract may reduce or eliminate fiduciary duties altogether. Members have no inspection rights where the information sits on an open blockchain. The articles must carry a conspicuous “NOTICE OF RESTRICTIONS ON DUTIES AND TRANSFERS” warning that member rights differ materially from those of an ordinary limited liability company. And the statute dissolves a DAO for failure to approve any proposal or take any action for a period of one year — a legislature writing an apathy clause into a corporate form, which tells you what the drafters expected the failure mode to be.

Established Four American states have such statutes and they differ in ways that matter.

Established Utah's provision is the one to read twice. A statute that limits liability generally and then restores it for members who vote a particular way has made the vote itself the liability trigger, which is the same doctrine the federal courts arrived at independently. Taken together with the Ooki and Lido holdings, the effect is a legal regime in which participating in governance is the act that exposes you. This brief's testable prediction, flagged as inference rather than finding: measured turnout in token-weighted organisations should fall as that doctrine propagates, because the marginal voter now bears a legal cost that the abstainer does not.

Established The engineering of bounded-membership systems looks nothing like this and is worth stating for contrast. Python's Steering Council model, accepted 17 December 2018, specifies five members elected by the core team after each feature release, each core member voting for zero to five nominees with the top five elected. Its powers are broad — accept or reject proposals, enforce the code of conduct, manage project assets, delegate authority — and its stated design philosophy is explicitly minimalist: “Minimum Viable Governance”, an instruction that the council “should look for ways to use these powers as little as possible”, trust and consensus-seeking in preference to rules, and a deliberate choice of “boring” models proven in other projects over experimental ones. There is no mechanism, no cryptography, and no attack surface. The whole apparatus is a page of text and a list of names, and it has governed a language with tens of millions of users through a founder succession without incident.

The nearest neighbour on this map is Future Federalism, and the boundary between them is the sharpest in this category. That brief owns jurisdictions: equalisation formulas, devolution settlements, subnational debt, secession, the tier structure of states. This brief owns associations: membership rather than residence, voluntary rather than compulsory. Read together they make an argument neither makes alone, because the two literatures converge on the same variable from opposite directions. The federalism evidence finds that the sign of decentralisation's effect is set by whether some institution disciplines the decentralised actor — national party discipline, a local media market — rather than by proximity. This brief finds that the sign is set by whether the membership boundary can be bought. Both are statements that decentralisation is not the operative variable, and that what matters is the accountability structure wrapped around it.

Collective Intelligence studies aggregation where this brief studies its binding form: the question there is whether a group can produce a better estimate, the question here is whether a group can produce a decision it must then live with. Future Democracies meets sortition and quadratic mechanisms from the constitutional side, and inherits the Colorado finding — that anonymity and transparency law are structurally incompatible — as a design constraint rather than a curiosity. AI-Assisted Governance shares the finding that a transparent record is not the same as an accountable process, which is this brief's central methodological point stated in a different domain.

Outside the map: institutional economics and the commons literature, which supplies the only strong quantitative evidence here; the sociology of organisations, where Michels' iron law has been tested on complete population data and survived; mechanism design; cooperative law; software engineering research, which turns out to hold the best natural experiment in the subject; and internet governance scholarship, which this brief could not adequately consult and says so.

Established The defining institutional fact is that the courts got to the legal-personality question before the legislatures did, and reached the opposite answer. Four American states wrote bespoke statutory forms — Vermont's blockchain-based limited liability company in 2018, Wyoming's DAO LLC in 2021, Tennessee's equivalent, and Utah's limited liability decentralized autonomous organization in June 2023. Meanwhile the federal courts held that an unregistered organisation is an unincorporated association capable of being sued, and can be pleaded as a general partnership. So the institutional landscape is not “no legal form exists”; it is “the default legal form is the worst available one, and the alternatives require a jurisdiction”. Utah's statute limits member liability to on-chain contributions but restores proportional liability for members who vote against compliance, which means a legislature and two federal courts independently converged on making the vote the liability trigger.

Established Wyoming's statute contains an institutional admission that deserves more attention than it gets. It dissolves a DAO for failure to approve any proposal or take any action for a period of one year. A legislature drafting a corporate form for these organisations wrote an apathy clause into it, which is a statement about what the drafters expected to happen. The same statute permits fiduciary duties to be reduced or eliminated by the articles or the smart contract, removes inspection rights where the information is on-chain, and requires a conspicuous notice that member rights differ materially from an ordinary company's. Read together: the form is available, and the state has warned you in capital letters.

Established The institution that has worked longest is the one with the least institutional apparatus. The IETF has no membership, which is why it cannot vote, which is why it adjudicates instead: a chair determines whether technical objections have been answered, subject to appeal, under a written norm that is explicit about its own fragility — “failing to be vigilant about sticking to the principles makes it increasingly hard to stick to them in the future.” No treasury, no token, no legal wrapper, decades of output. Python's Steering Council is the same shape at a smaller scale: five people, elected by a bounded electorate of about a hundred, instructed to use their powers as little as possible.

Established ICANN is the institution that scaled the model and the evidence on it is its own. Its 2020 self-assessment identifies insufficient prioritisation, over-wide scoping producing delay, duplication and implementation failure, an absence of incentives to compromise, volunteer burnout, chairs contending with capture tactics, participants winning by delay, silos and distrust, and a zero-sum approach to policy-making. It contains no operating statistics whatsoever. This brief has no independent academic evaluation of ICANN — the one it sought was inaccessible — so every criticism of ICANN here is ICANN's own, which is the weakest possible standing for a critical claim and makes the concessions more rather than less notable.

Frontier The institution that does not exist is an independent auditor of distributed governance. There is no body that scores an organisation against the commons design principles, no registry of governance concentration, no equivalent of a proxy-advisory firm for token voting that is not itself a market participant. The delegate-ranking interfaces that shape delegation are run by governance-platform companies whose incentive is engagement. The one genuinely independent measurement infrastructure is academic, unfunded for this purpose, and publishes on a two-year lag — which is why this brief's concentration figures run to June 2025 and its population study to 2024, and why its turnout claims should be read as dated rather than current.

Frontier The institution that is quietly doing the work is the volunteer moderator. Federated social networks devolved moderation faster and further than any other function: 62 labelling services on one network, community labellers producing 88.7% of labels. The burden they absorb is as concentrated as everything else — on one federated network a single instance produced 23.33% of all toxic posts and the top three about 31% — and it is carried by unpaid people with no accountable monitoring, which is the design principle the commons evidence treats as necessary and which is absent here. The most successful decentralisation in this brief is the decentralisation of an unfunded liability.

Established And the largest institution in the subject is a cooperative federation. More than 70,000 people, €11.2 billion of sales, roughly 30,100 worker-members, largely conventional subsidiaries abroad, and a demonstrated willingness to let a founding member fail. Anyone reasoning about distributed ownership at scale is reasoning about this organisation whether they intend to or not, and the honest reading of its record is that it survived by acquiring the capacities — delegated authority, professional management, the ability to refuse — that hierarchy is said to be uniquely for.

Established The core ethical difficulty is that token weighting makes governance rights purchasable, which converts a political question into a market one by construction. The commons literature this field cites treats a defensible membership boundary as necessary for success; token governance has none, and the concentration figures are the arithmetic consequence. The DAO's top 100 holders held just over 46% within three weeks of launch. That is not a failure of implementation and it cannot be fixed downstream.

Established The second is a wealth qualification implemented in transaction costs rather than in law. On a median contribution of $206.26, a round trip at roughly $60 each way consumes about 58% of the stake; roughly $23 million of a $47 million raise remained unrefunded weeks after the event because claiming cost more than a quarter of what it returned. Systems that are open in principle and priced out of reach in practice have a franchise restriction, and describing them as open is a claim about rules rather than about who participates.

Frontier The third is accountability without recourse. When a valid vote transfers a fifth of a treasury to the people who proposed it, dissenters have no appeal — no court, no regulator, no membership to expel anyone from. The observed remedies are to exit, to fork, or to abolish the governance system, and each imposes the cost on people who did nothing wrong. The emerging legal answer is worse in a different way: general-partnership liability makes the dissenter jointly liable for the outcome they voted against.

Established The interested-party problem is acute here because most of the reporting is done by the organisations reporting on themselves. Quadratic-funding round statistics come from the grant-maker, and its anti-fraud evaluations from a contractor that grant-maker pays. The quadratic-voting field-trial success claims come from the foundation that exists to promote quadratic voting, which publishes no vote totals or rankings, making them unverifiable. The one peer-reviewed futarchy paper is written by authors who disclose holdings in the organisations they study. Cooperative survival rates come from a sector federation. Token-burn analysis comes from a firm serving the sector. ICANN's only self-assessment is ICANN's. This brief marks every such source in its reading list, states what it relies on each for and what it does not, and rests every concentration figure on academics with no protocol affiliation, on complete-population repository studies, on a regulator's investigation or on court judgments.

Established The independent measurements are the computational ones, and that is the real methodological gift of this subject. Where the data are on-chain or in a git history, the evidence is unusually good and reproducible by anyone. Where they are off-chain — the filter, the deliberation, the outcome quality — the evidence is as bad as anywhere else and mostly self-reported. A brief that quoted only the good half would be describing a different subject.

Established The cooperative case raises the sharpest version, because it is not hypothetical. A democratic federation decided, through its own procedures, to let a founding member cooperative fail rather than subsidise it, and some 5,700 people lost their employer. That is the model working as designed. Distributed governance is not a way of avoiding hard decisions; it is a way of distributing responsibility for them, which is a different and less comfortable claim.

Four questions this brief does not resolve and states as obligations. Whether the cooperative survival advantage is structural or a selection artefact — the gap is large and replicated across five jurisdictions and nobody has identified it. Whether platform cooperativism has any measurable record at all: this brief could not open the systematic literature review or two of the three case-comparison studies it sought, and can therefore name Stocksy, Fairbnb and Resonate as examples while offering no verified figure for the number of platform cooperatives, their revenue, their membership or a single outcome comparison against an investor-owned platform. Whether standards-body participation is dominated by a small number of large firms' employees — a widely voiced concern for which this brief could verify no figure, and which therefore appears here and nowhere else. And whether measured turnout falls as the general-partnership doctrine propagates, which is answerable from the public record by anyone who cares to look.

Established The civilisational finding is that the historically successful form of distributed governance is the demanding one, and that its demand is a boundary. Commons institutions that lasted centuries did so with clear membership, monitors, graduated sanctions and conflict-resolution machinery — a full institutional stack, locally adapted, with the quantitative work showing that near-complete satisfaction of the principles is what predicts success. The contemporary forms that market themselves as successors adopted the branding and skipped the stack, and the one thing they most conspicuously skipped is the boundary. The general principle this case illustrates is not about blockchains: an association can govern itself without a centre for as long as it controls who is in it, and no longer.

Established The second civilisational finding is that the iron law is now an empirical result and not a slogan, which changes what can honestly be promised. On 683 organisations with complete population data, in a setting with no capital and no exclusive membership, formal authority concentrated in a founding cohort as the population grew, and administrative power was increasingly used against experienced participants. Michels wrote in 1911 about political parties; the claim now has an out-of-sample test on a substrate he could not have imagined, and it passed. Any proposal to govern something important without a centre must now carry an account of why this case escapes a regularity that has held in parties, unions, cooperatives, wikis and token organisations. None of the proposals currently in circulation carries one.

Frontier What this movement has genuinely contributed is measurability, and that contribution is real and undervalued. For the first time it is possible to compute exactly how concentrated a governance system is, how many people actually voted, whether the powerful exercised their power, and who really writes the code. Those numbers are worse than the rhetoric and better than the cynicism, and both corrections come from the same data. Political science has never had complete-population data on 683 organisations' internal authority structures, or the full authorship history of the software running the world's infrastructure. That instrument outlives whatever happens to the tokens.

Handwave Claims that distributed governance is ready to administer anything at civilisational scale describe no deployment that exists. The largest working non-hierarchical organisation is a cooperative federation where fewer than half the workforce are members, whose most consequential democratic decision in fifty years was to let a founding member fail. The largest federated social network has 98.9% of its identities under one provider. The most durable protocol body works by having a chair decide when objections have been answered. Everything in this brief that has worked for decades has a small accountable hierarchy at the decision point, and the honest civilisational reading is that this may be the discovery rather than the disappointment.

Established What already happened, because the chronology usually starts too late. Commons institutions of the kind the design-principle literature documents have governed irrigation districts, inshore fisheries and forest user groups for centuries. The IETF's consensus practice predates the web and was written down as an Informational RFC in June 2014, decades after it began. Mondragón was founded in 1956. The DAO launched and failed inside eleven weeks in 2016. Wyoming's DAO statute passed in 2021, Utah's in June 2023. The Ooki default judgment landed on 8 June 2023. Python's Steering Council was elected in February 2019. None of the important dates in this brief are in the future.

Frontier Now to about 2030: the legal question resolves, and it resolves through litigation rather than legislation. Whether governance participation carries general-partnership liability will be decided in United States federal courts on the current docket, not by state statutes drafted for the purpose. The statutory forms will continue to exist and will continue to be used by organisations willing to accept a jurisdiction; the ones unwilling will continue to be unincorporated associations, which is a legal status, not an absence of one.

Frontier The same interval: timelocks, bytecode review and delegate-concentration disclosure become standard, because each was learned from a specific loss. This requires no research and no coordination; it is what insurers and auditors will ask for.

Speculative Mid-2030s: the plausible split. Token governance settles as a treasury-management and protocol-parameter mechanism — narrow, technical, low-stakes per decision — while anything with distributional consequences reacquires a hierarchy, as every observed failure has forced. The governance-token-as-asset question is already resolving in that direction: the flagship token conveyed governance and nothing else for about five years before cash flow was attached to it in late 2025.

Speculative Longer: commons-style governance of genuinely shared resources continues to work where it always has, at the scale it always has, with the institutional stack it has always required. Nothing in the evidence suggests that scale is about to change, and the one serious attempt to extend the principles to global commons concluded that boundary definition — the load-bearing principle — is not a meaningful exercise at that scale.

Handwave Any date attached to distributed governance administering something at civilisational scale. There is no deployment to extrapolate from. The largest working non-hierarchical organisation in the world is a cooperative federation in which fewer than half the workforce are members, and its most consequential democratic decision was to let a founding member fail. Dates in this category are not forecasts and this brief declines to supply one.

Handwave Whether anyone applies the commons design principles as an actual specification. Nobody has. The design principles have been cited for four decades and treated as a description of what survived rather than as a buildable requirement, and the one setting where the boundary principle was tested by accident — a purchasable electorate against a bounded one — produced an order-of-magnitude turnout gap that nobody designed for. Predicting when a field will start using its own best result as an engineering constraint is not forecasting.

Handwave Anything beyond that horizon. The underlying problem — how a group binds itself when nobody is empowered to enforce the binding — is among the oldest in political thought and has no technical solution. Cryptography can make a vote unforgeable and a treasury unspendable without a quorum. It cannot make a quorum care, and the measured record of this field is very largely a record of quorums not caring.

• Depends on Nothing on this map. The cryptography works, the voting contracts work, the commons literature is fifty years old and the standards bodies predate the web — nothing here waits on a result produced by another brief. The constraints are institutional, legal and behavioural, and they are recorded below. Future Federalism is a neighbour rather than a dependency: it owns territorial multi-level government, this brief owns governance of associations rather than jurisdictions.
• Requires (not on this map) A membership boundary that is not simply purchasable — the principle the commons evidence treats as most load-bearing, the one token governance is constituted to violate, and the variable that separates a 72% turnout from a 5% one. Graduated sanctions and conflict-resolution machinery, two further design principles the quantitative work finds necessary and which token-weighted governance almost universally lacks. Accountable monitoring separate from the participants being monitored, which in federated networks currently means unpaid volunteers carrying a burden as concentrated as the rest of the system. A legal wrapper able to hold a treasury and bear liability, now supplied at the price of accepting a jurisdiction. And settlement cheap enough that a round trip does not consume a majority of the median stake. All six are institutional, legal or economic capabilities rather than discoveries, and five of the six are specified by a literature this field cites and does not implement.
• Enables Collective decisions binding on participants without a central authority to enforce them. No typed enabling edge is claimed: the briefs that assume coordination assume it generically rather than depending on this route to it, and the honest statement is that nothing on this map is currently waiting on distributed governance to work.
• Adjacent Institutional economics and the commons literature, which supplies the strongest quantitative evidence in this subject; the sociology of organisations, where the iron law of oligarchy is now a tested empirical claim rather than a slogan; mechanism design; cooperative and corporate law; and within this map Collective Intelligence, Future Democracies, AI-Assisted Governance and Future Federalism.

“DAO concentration is roughly Gini 0.9-plus.” Established Half right, and the missing half is the dispersion. At population scale across 100 organisations the mean holder Gini is 0.794 with a standard deviation of 0.275, and the mean Nakamoto coefficient is 35.9 with a standard deviation of 120.4 across a range of 1 to 1,046. A standard deviation more than three times the mean describes a bimodal population, not a concentrated one. The flagship figures — Gini above 0.99 for Aave, Compound, ENS and Uniswap — are correct for the flagships and wrong as a population statement; the population mean is correct as a population statement and useless for anything else. The honest version is that the most-studied protocols are the most concentrated ones, and 42% of organisations hold under $1 million, so most of the population is not governing anything of consequence.

“Extreme concentration means capture.” Established It does not, and nearly every account conflates them. Gini coefficients above 0.98 are real; so is the finding that large delegates voted against the wider community on 6 of 84 proposals in one protocol and 1 of 10 in another. Reporting only the concentration is reporting half the measurement. Established But near-universal pass rates are not evidence of consensus either: contested questions are filtered out off-chain before reaching a vote, so the measured object is ratification rather than decision, and an on-chain pass rate near 100% measures the filter.

“The legal-personality problem for DAOs is unsolved.” Established It has been solved, against the organisations. An unregistered DAO is an unincorporated association capable of being sued — Ooki DAO, 8 June 2023, a $643,542 civil penalty and a permanent injunction, with service of process effected through the organisation's own help chat box and web forum — and can be pleaded as a general partnership with unlimited joint and several liability, as in Samuels v. Lido DAO. Both holdings turn on the fact that token holders vote. The statutory alternatives require registration, a registered agent, and a conspicuous notice that member rights differ materially from an ordinary limited liability company — that is, they require the organisation to acquire the territorial legal anchoring it was constituted to escape. A liability regime that penalises participation in governance is an odd foundation for a governance movement.

“Governance tokens confer governance.” Established For the flagship case they conferred governance and nothing else for about five years, from launch in 2020 until cash flow was attached in late 2025, and the market priced them throughout on expectations of a future in which they would convey something else. When the change came, the vote was 99% in favour with virtually no opposition and more than 69 million tokens voting yes, retiring 100 million tokens from treasury — 10.1% of the original billion, worth roughly $940 million. In the first twelve days after activation about 100.17 million tokens (roughly $557 million) were burned against cumulative protocol fees of about $0.8 million, annualising to $26–27 million against a market value of $5.4 billion — a revenue multiple of roughly 207× — with an ongoing burn of 4 to 5 million tokens a year running against a 20 million-token annual growth budget. Read that vote through this brief's own framework: near-unanimous, uncontested, proposed by the beneficiaries. Ratification, not decision.

“Ostrom is a licence for governance without structure.” Established The design principles specify boundaries, monitors, graduated sanctions, conflict resolution and nested enterprises, and the predictive work finds that success requires satisfying ten or eleven together, with fisheries requiring all eleven. That is a demanding formal institution. The literature is evidence against central state control of commons, and it is routinely cited by organisations that satisfy perhaps three of the eleven.

“And since Ostrom shows communities can govern commons, commons methods scale.” Established The extension attempt is on the record and it names what breaks. For global commons, defining boundaries for resources and appropriators “is not a meaningful exercise”; direct participation by most users is infeasible and must be weakened to participation by interested and affected parties; monitoring independence is essential and structurally hardest, because users fund the monitoring; local autonomy must yield to higher-level restriction; and learning from direct experience is not feasible, because the errors are irreversible. The conclusion is that the principles require modification and extension, not that they fail — but the evidence base was collected on irrigation districts, inshore fisheries and forest user groups of tens to low thousands of members, and it does not reach millions of anonymous participants.

“Open source proves distributed decision-making works.” Established It proves distributed contribution works and shows that decision-making is not distributed at all. 57% of 1,932 popular repositories have a truck factor of one. An independent study of the top 1,000 repositories by stars puts half at a bus factor of two or less, only 10% at six or above, with no correlation between popularity and bus factor. 16% of projects were abandoned when all core maintainers left, 66% of those in truck-factor-one projects; 41% recovered, in 86% of cases through one replacement person. The constitution and the production structure are different objects and only one of them is distributed.

“The benevolent-dictator-to-foundation transition decentralised open-source governance.” Established In the best-documented case it moved formal authority and left informal authority where it was. The governing document explicitly rejects the benevolent-dictator model and explicitly adopts “Minimum Viable Governance”, instructing the council to “look for ways to use these powers as little as possible”. The first election, 69 of 96 eligible voters, returned the former benevolent dictator and four other long-standing figures on what the specialist coverage called complete landslide majorities, with the outcome read as choosing the status quo. Michels would have predicted the result and the vote-counting rule that produced it was criticised at the time for making it more likely.

“Rough consensus is consensus.” Established It is adjudication. “Rough consensus is achieved when all issues are addressed, but not necessarily accommodated”, and the organising principle is that “lack of disagreement is more important than agreement”. Voting is rejected partly because “since the IETF is not a membership organization, it's nearly impossible to figure out who would get a vote”, and partly because it permits vote-stuffing and majority tyranny. The two symmetric statements are the substance: “One hundred people for and five people against might not be rough consensus” where the five have an unanswered technical objection, and “Five people for and one hundred people against might still be rough consensus” where the hundred have no technical substance. A chair decides whether an objection has been answered. The most durable non-hierarchical governance institution in the record works by installing a small accountable authority at the decision point and constraining it with a norm rather than a rule.

“Federated networks give users an exit, so they are governed by exit.” Established The exit exists and is not taken. 96% of Mastodon users are on 25% of the largest instances. 98.9% of Bluesky handles sit under bsky.social, with 57,202 on other domains and half of those through four registrars. 95.8% of that network's 40,398 feed generators are hosted by three platforms and 85.86% by one; the relay has no documented competitor. The component that has dispersed is moderation — 62 labelling services, community labellers producing 88.7% of labels — which is the expensive, unglamorous, legally exposed function.

“Multistakeholder governance is a mature alternative.” Established ICANN's own 2020 self-assessment names insufficient prioritisation, scoping so wide it produced “endless discussions”, duplication and implementation failures, an absence of incentives for stakeholders to compromise, volunteer burnout, working-group chairs contending with “capture tactics”, participants winning by delay, a silo mentality with a perceived lack of trust, and a zero-sum approach to policy-making. It contains no figures at all on how long its processes take, how many volunteers run them, or how many working groups exist. Twenty years in, the flagship multistakeholder body does not publish the basic operating statistics of its own governance model, and that absence is the finding.

“Cooperatives fail because worker ownership cannot compete.” Frontier The survival evidence points the other way, though it is not identified. In the United Kingdom 2011 cohort, five-year survival was 80% for co-operatives against 44% for all companies, with one-year 96% against 93% and three-year 87% against 61%; comparable gaps are reported for Portugal (75 against 40), France (81–89 against 40–50), Belgium (80 against 68) and Alberta (84.6 against 45). The sector federation reporting this is an interested party and says so, noting that the underlying studies “deliberately select and refine data to analyse specific sectors, co-operative ownership types, time intervals, or precise geographical areas” and that further statistical work would be needed to establish whether the gap is structural. It is consistent with cooperatives being more resilient and equally consistent with cooperatives forming in different sectors, at different sizes, under different selection. The gap is large, replicated across five jurisdictions, and unidentified.

“Distributed governance's characteristic failure is capture by a small group.” Frontier On the record assembled here the characteristic failure is abdication. Fagor was not captured: its assembly was too weakly informed to challenge management and too slow to close a unit losing more than €14 million a year for fifteen years, and turned out in numbers only when its own pay was at stake. The remedies its own analysts propose are delegate assemblies above 3,000 members, growth by spinoff rather than by growing one large cooperative, independent professional directors and mandatory council training — that is, less direct democracy. In the wiki data the mechanism is the same in a setting with no capital and no boundary: a founding cohort accumulates formal authority faster than the population grows, and uses it increasingly against experienced contributors. Power that formally belongs to everyone and is exercised by almost nobody flows to whoever is present.