Note 73: The Multi-Stakeholder Mirage --How the Multi-Stakeholder Model Turned Attendance Into Mandate

The multi-stakeholder model began as a compromise, not a constitution. It gave the Internet a way to avoid state monopoly, corporate monopoly and intergovernmental bureaucracy at the same time. It allowed governments, companies, engineers, civil society, academics, registries and users to appear inside one political picture.

The picture was useful because it was vague. It sounded broader than private control, safer than government rule, more open than bureaucracy and more democratic than technical administration. But it concealed the only question that mattered. Who, exactly, had the power to authorize binding decisions?

That is the multi-stakeholder mirage. The stakeholders are real; the mandate is not. A stakeholder is affected by a decision, while a principal authorizes one. A model that cannot keep those categories separate does not solve legitimacy. It makes legitimacy appear where no authorization exists.

This is not an argument against participation. Participation can be evidence, expertise, warning, objection and technical discipline. It becomes dangerous when it is inflated into authority over absent parties. The affected may speak, but they do not thereby acquire the right to bind the party that bears the loss.

The three borrowed sources

The model survives because it borrows from three histories and treats them as one constitutional source. It borrows the prestige of the IETF, the institutional settlement of ICANN, and the diplomatic vocabulary of WSIS. Each source is real. None supplies the mandate now claimed in its name.

The IETF tradition was a discipline of technical work, not a theory of political representation. RFC 7282 explains rough consensus in opposition to kings, presidents and voting, and treats implementation experience as a check on abstract decision-making. That is engineering discipline. It is not a license for a room to govern assets, capital markets, national infrastructure or operator continuity.

ICANN solved a different problem. The 1998 White Paper pointed toward stability, competition, private bottom-up coordination and representation for the DNS transition. That bargain helped avoid direct state operation of key Internet coordination functions. It did not create a world electorate, a user demos or a general public mandate over the Internet.

WSIS supplied the broadest political vocabulary. The Tunis Agenda placed governments, the private sector, civil society and technical actors inside one Internet-governance frame. But it also made the IGF non-binding, with no oversight function and no involvement in day-to-day or technical operations of the Internet. A microphone was not sovereignty.

The modern model fused these histories into one legitimacy story. When it needed technical prestige, it invoked rough consensus. When it needed global legitimacy, it invoked WSIS. When it needed insulation from government, it invoked private coordination. When it was challenged on mandate, it invoked “the community.”

The strongest defence

The strongest defence of the multi-stakeholder model is that it helped prevent intergovernmental capture. That defence has force. A government-led or treaty-based Internet-control system could have been slower, more censorial and more vulnerable to geopolitical bargaining. NTIA made support for the multistakeholder model a condition of the IANA stewardship transition and said it would not accept a government-led or intergovernmental replacement.

That historical defence should be admitted, not brushed aside. The Internet did need a way to avoid becoming a treaty bureaucracy or a single-state instrument. Private-sector coordination, technical autonomy and open participation helped preserve speed and flexibility at a crucial stage. The model was not born useless.

But avoiding one failure does not justify another. The danger of government monopoly does not make undefined community mandate lawful. The risk of treaty control does not make private chokepoint authority accountable. Participation is better than closed bureaucracy, but it is not the same as authorization.

The real lesson is narrower. The early Internet settlement avoided direct intergovernmental capture, but it did not build hard limits on the private chokepoints that replaced it. It confused “not government” with “legitimate.” It confused “open participation” with “lawful representation.” It confused “community input” with “principal authority.”

The arithmetic of the room

The numbers are not merely small. They are wrong for the constitutional work assigned to them. ITU estimates that about six billion people are online in 2025, while 2.2 billion remain offline. That is the scale against which claims of user representation must be tested.

ICANN’s At-Large Advisory Committee has fifteen members and is described in ICANN’s bylaws as the primary organizational home within ICANN for individual Internet users. Fifteen people can advise on user interests. They can organize outreach, submit comments and identify user-facing concerns. They cannot represent six billion users.

ICANN’s corporate layer is more honest than the rhetoric around it. ICANN has sixteen voting directors, and its legal architecture does not operate as a global membership democracy. ICANN’s bylaws state that ICANN shall not have members, and that the Empowered Community is not a member of ICANN. That may be workable corporate governance. It is not popular sovereignty.

ICANN85 makes the participation gap visible. The official report states that ICANN85 had 2,195 attendees, with 1,517 in person and 678 virtual, and says ICANN Public Meetings are a central pillar of ICANN’s multistakeholder model. The same report records average in-person attendance per session of 47, based on mid-session room counts. A room of 47 people may be useful. It is not a mandate.

The compression ratio is absurd if treated as representation. Fifteen ALAC members are made to sound like the user layer. Sixteen voting directors remain the corporate board. Forty-seven people can become an average physical session. The multi-stakeholder model turns scale mismatch into moral vocabulary.

The permanent forum problem

The IGF shows the same distinction at global scale. It can convene governments, companies, engineers, civil society and academics. It can surface issues, widen access and build capacity. It can be useful precisely because it is a forum.

The IGF’s recent statistics show a substantial discussion layer. In 2024, the Riyadh IGF recorded over 10,143 total participants, with 7,343 onsite and over 2,800 online. In 2025, Lillestrøm recorded over 9,435 total participants, with 3,435 onsite and over 6,000 online. These are large meetings. They are still not a demos.

WSIS+20 made that forum permanent. The IGF says the UN General Assembly confirmed it as a permanent UN forum under Resolution 80/173, and the 2026 IGF will be the first meeting after that decision. That is institutionally important. It does not solve the execution-layer problem.

A permanent forum is still a forum. It may make dialogue durable, but it does not create portability, failover, liability symmetry or operator authorization. Governments may feel included because they have a standing microphone. But a microphone is not a lever.

This is the WSIS+20 alibi. The discussion layer becomes permanent while the operational layer remains elsewhere. Names, numbers, registry records, RPKI, reverse DNS, transfer compatibility and continuity mechanisms remain in private or quasi-private structures. Governments receive a forum. Administrators keep the switch.

The missing ledger

The strongest evidence is not only what the data show. It is also what the data cannot show. A model that constantly invokes openness should be able to produce comparable long-run data on attendance, voting, remote participation, mailing-list activity and decision participation. Much of the system cannot.

ICANN produces relatively clear meeting statistics. APNIC60 also provides a useful snapshot: 451 in-person attendees, 60 online-only participants, 40 economies represented, 127 APNIC member organizations represented and three policy proposals reaching consensus. That is useful operational data. It is not regional authorization.

For the wider RIR system, the public record is less standardized. Some sources provide meeting-week totals rather than policy-forum totals. Some provide attendee lists that function only as lower bounds. Some combine events, hide remote participation or fail to publish the denominator needed to test representativeness.

This strengthens the critique. If legitimacy rests on open participation, participation must be auditable. Where the table says “unspecified,” the representative claim should become weaker, not stronger. A legitimacy model that cannot show its denominator should not be treated as a constitutional model.

Votes without mandate

Formal voting is better than loose consensus. It creates a clearer internal act and a more reviewable record. But voting still does not automatically turn a stakeholder process into a principal structure. A vote can authorize what the voting body has legal authority to decide; it cannot authorize everything later described as community policy.

RIPE NCC illustrates the limit. Its May 2025 GM turnout fell to 5.3%, with 1,039 actual votes from 19,713 eligible members, according to RIPE NCC’s own post-election analysis. Those votes matter inside the association. They do not become a public mandate over the economic destiny of number resources.

ARIN’s figures appear stronger, but they raise a different problem. ARIN’s 2023 election page reported 959 votes from 6,197 eligible organizations. ARIN’s 2024 official page reports 862 votes and 43.91% turnout in one section, while its voter-statistics section says 959 General Members cast ballots from 1,963 eligible organizations. That inconsistency should be disclosed, not smoothed away.

The point is not that RIPE or ARIN has no internal legitimacy. The point is scope. Association voting may authorize association decisions. It does not convert a registry into a public-law governor over scarce assets, operator continuity or national infrastructure risk.

Lists are not legislatures

Mailing lists remain useful. They preserve objections, reduce travel costs and allow asynchronous technical discussion. They are often better than rooms because they leave archives. But they are not legislatures.

APNIC’s Policy SIG charter covers policies for number-resource management and related services such as allocation, recovery, transfer, WHOIS, reverse DNS and RPKI. Its public activity pages are useful as snapshots, not as constitutional proof. Even where list data exist, they show a channel of participation. They do not show regional consent.

RIPE’s own mailing-list history is more nuanced. It shows that RIPE mailing lists remain part of the community’s working infrastructure, while activity has changed over time with issue cycles such as IPv4 exhaustion and transfer disputes. That is useful sociological evidence. It does not convert list activity into legal representation.

The deeper problem is role confusion. The RIR system mixes corporate representatives, database contacts, technical contacts, consultants, staff, proxies, policy participants and regular speakers. Only some can bind a legal principal, and only within scope. A database contact is not a corporate power of attorney.

The operator disappeared

The missing principal is the operator. The operator signs customer contracts, deploys routers, pays transit, announces prefixes, maintains abuse desks, answers regulators and carries uptime obligations. When a registry policy delays a transfer, impairs RPKI, suppresses asset value or creates registry uncertainty, the operator bears the cost.

The multi-stakeholder model treats the operator as one voice among many. That may be tolerable for narrow technical advice. It is not tolerable when the subject is revocation, transferability, leasing, capital value, customer geography, security assertions or continuity of live networks. Those questions reach the operator’s balance sheet and customer base.

This is where the wider Lu Heng argument matters. The operator is the central actor in Running-Code Primacy, because running networks are not symbolic participants in a room. They are the system that the registry layer was originally supposed to serve. The operator is also why LARUS treats IPv4 not as a moral token but as operational infrastructure.

Civil society may raise valid concerns. Governments may regulate under public law. Engineers may identify real technical invariants. Consultants may explain market dynamics. None of them thereby becomes the operator. A policy room may advise operators, warn them, criticize them and supply evidence. It cannot become them.

When identifiers became capital

The old registry model was designed for a low-value world. Number resources looked abundant, clerical and low-conflict. Informal community process seemed adequate because the thing being coordinated looked like an address book. That premise no longer holds.

IPv4 scarcity changed the object. Number resources became scarce, leased, transferred, financed, litigated and operationally embedded. The database record began to affect capital value, transferability, customer continuity, routing credibility and security assertions. A clerical entry became a gate to economic use.

That change is developed across the notes on registry power and liability, number resources as non-political assets, and thick governance as double extraction. The point is not merely that IPv4 has a price. The point is that an administrative record now sits above capital, continuity and operator reliance.

That change should have narrowed registry authority. Instead the language expanded. Stewardship became control over scarcity. Community became a claimed principal. Region became political property. Need became capital allocation.

This is the economic failure of the multi-stakeholder model. It kept the legitimacy language of a consultative technical world after the governed object had become asset-grade infrastructure. A model built for discussion cannot govern capital without principal authorization, liability and exit.

The capital discount

Economics matters because price reveals power. A chokepoint does not need formal ownership to extract value. It needs dependency. If a holder cannot leave, the registry’s control over records becomes leverage.

The costs are visible and invisible. Visible costs include fees, documentation burdens, compliance work and delay. Invisible costs include reduced liquidity, weaker collateral value, uncertainty over leasing, transfer friction, legal risk and the possibility that registry discretion will impair assets without bearing proportional loss. The invisible costs are often larger because they are priced as uncertainty.

Once an input becomes capital, governance of that input must change. The system needs clear rights, low transaction costs, predictable recording, reliable transfer, dispute isolation and liability symmetry. If those elements are missing, the asset is discounted. That discount is paid by operators, customers and regions.

This is why the multi-stakeholder mirage is not merely a legitimacy problem. It is a cost-of-capital problem. It delays transactions while claiming stewardship. It suppresses liquidity while claiming equality. It destroys price discovery while claiming public interest.

The poverty claim

Restrictions are often defended in the name of poorer regions and smaller networks. The argument says transfer limits, needs tests and leasing restrictions protect weak actors from market extraction. It sounds moral. It is economically backwards.

Scarcity controls do not create more IPv4. They create illiquidity, raise transaction costs, reduce collateral value, discourage inbound supply and reward insiders who can navigate process. Rich networks can hire lawyers, structure around restrictions and survive delay. Poor networks cannot.

This is the logic of The Poverty Penalty. Poor networks need access, liquidity, financing, transparent transfer records and low transaction costs. They do not need a gatekeeper speaking moral language while preserving discretion. Price can be compared, budgeted and financed. Discretion cannot.

A bad market can harm the poor. A discretionary anti-market regime can harm them more. It gives them fewer exits, less information and greater dependence on institutional favor. That is not equality; it is a poverty penalty.

The user as ghost

The end user is invoked constantly and represented almost never. Most users do not know the meetings exist, do not vote in them, do not authorize speakers and do not understand transfer policy, registry contracts, RPKI or address scarcity. They experience outcomes. Service works or fails. Access becomes cheaper or more expensive. Networks remain reachable or they do not.

That makes the user politically useful. Every institution can claim to speak for the user because the user is absent from the authorization chain. Boards, registries, companies, governments and civil-society groups can all place the user inside their own sentence. The user becomes a ghost that legitimizes institutional speech.

The proper test is not who says “user.” It is whether the rule lowers access cost, improves reliability, preserves security, increases portability, reduces lock-in and protects the networks users depend on. Outcomes matter more than invocations. A model that invokes users without letting them authorize anything is not representative; it is symbolic.

This is also why a reality layer matters. BTW.Media and the note on why BTW.Media exists are useful here because the first task is not advocacy language. It is making hidden institutional structure visible enough that users, operators, governments and markets can understand what is being done in their name.

The technical alibi

The technical community’s original legitimacy came from competence and restraint. It solved problems of protocol, naming, numbering, routing, security and interoperability. It persuaded because operators could test results. The authority was practical, not sovereign.

The multi-stakeholder model borrowed that legitimacy. It kept the language of consensus while weakening the discipline of running code. A rule could become binding not because operators voluntarily adopted it, but because registry records, contracts, transfer recognition or security assertions made non-compliance expensive.

That is the Running-Code Betrayal. Running code once disciplined consensus. Consensus later justified the multi-stakeholder model. The model then claimed authority to override running code. The child claimed power over the parent.

Technical experts should remain central to technical questions. They should define invariants, expose implementation risks and test deployment reality. They should not be turned into a substitute demos for asset governance. Technical competence is not a blank cheque for institutional enforcement.

The corporate layer tells the truth

Corporate law often speaks more honestly than governance rhetoric. ICANN has a board, bylaws, advisory structures and accountability mechanisms. It has no ordinary corporate members, and the Empowered Community is not a corporate member. That is a legal architecture, not a global electorate.

The same clarity should be applied to the RIR world. A registry can be a useful recordkeeper. A member association can make internal decisions. A policy community can develop advice. None of these facts creates sovereign authority over number resources.

The danger comes from combining legal modesty with rhetorical grandeur. The institution acts through boards, staff, contracts and policies. When challenged, it invokes community, region and multi-stakeholder legitimacy. Enforceable power travels through a soft legitimacy cloud.

That is the mirage in legal form. Everyone is invoked at the front end. A small institution acts at the back end. The principal remains empty.

The state’s bad bargain

Governments accepted the multi-stakeholder bargain because the alternatives looked worse. A treaty-based Internet could have been slower, more censorial and more fragmented. A single-state Internet was politically unacceptable. Private coordination looked like the least dangerous route.

But governments kept the public downside. They remain responsible for economic disruption, security exposure, communications risk and political pressure when critical coordination systems fail. Yet key operational levers remain inside private or quasi-private structures: numbering, naming, registry records, RPKI, reverse DNS, transfer compatibility and continuity mechanisms.

That is Sovereignty Inversion. The state carries public responsibility while the private administrator holds operational leverage. The multi-stakeholder room then supplies the language that makes the inversion look consensual. A government gets a microphone while the gatekeeper keeps the switch.

The answer is not government takeover. States can censor, fragment and politicize infrastructure. But rejecting state monopoly does not validate private chokepoint sovereignty. The correct answer is thinner common layers, portability, auditability, liability, failover and replaceable administration.

The stress test

Recent RIR crises should be treated as stress tests, not as the center of the argument. Their value is diagnostic. They show what happens when registry discretion collides with operationally embedded resources, ordinary courts, missing portability and system-level attempts to preserve institutional independence. The lesson is structural, not personal.

Public coverage of the AFRINIC/Cloud Innovation dispute by BTW.Media is useful in that limited sense: not as the center of this article, but as evidence of how registry theory behaves when it touches live assets, courts, governance failure and system-level defensiveness. The wider conceptual point is developed in The Stability Fallacy and The Registry Continuity Fallacy. The argument is not that one registry crisis explains everything. It is that stress reveals what normal ritual conceals.

If a registry claims broad discretion over valuable resources while there is no hard portability, no clean failover, no proportional liability and no precise representation theory, conflict will not remain administrative. It will become existential. The registry sits where law, markets, routing, security and customer continuity meet. A small institution can therefore create a large blast radius.

The key distinction is continuity of the ledger versus continuity of the gatekeeper. Registry continuity requires accurate records, RDAP or WHOIS, reverse DNS, RPKI, transfer history, dispute metadata and operational continuity. It does not require institutional immortality. Protect the ledger, not the throne.

The transition layers

The answer cannot be only critique. A defective institutional order can still carry real operational dependencies. That is why the transition must be staged: visibility first, continuity next, coordinated protection next, and only then thinner post-RIR architecture. Romantic rupture is not an architecture.

BTW.Media belongs to the visibility layer. It makes registry-side risk legible outside the small procedural class that normally controls the vocabulary. A reality layer matters because a mirage survives only while the audience cannot see the machinery.

LARUS belongs to the continuity layer. Its relevance is not that it is another market actor, but that it treats IP resources as operational continuity assets rather than clerical residues. LARUS One develops that idea at the network-identity layer: the delivery provider may change, but the public network identity should not have to break. The corresponding note, On LARUS One, makes the economic point directly.

i.LEASE belongs to the execution layer. Ordinary brokerage treats registry risk as paperwork. The note on why i.LEASE exists treats brokerage as execution under registry-layer uncertainty. That is the correct frame once IPv4 is capital and registry discretion can become transaction risk.

NRS belongs to the coordinated-protection layer. The point is not to create a new sovereign above operators. It is to give resource holders a way to resist registry-side risk together rather than as isolated targets. The note on why NRS exists frames decentralization as systems engineering, not ideology.

The demotion

The fix is not to abolish participation. It is to demote it. Participation should be evidence, not authority. Consensus should be technical judgment, not sovereignty. Community should be description, not principal.

The common layer should contain only what must be common: uniqueness, proof of control, registry accuracy, contactability, transfer records, security assertions, dispute metadata, auditability, portability, failover and replacement paths. Everything else should stay closer to the operator, contract, market or public law. Commercial use, leasing, customer geography, financing, routing practice and business model do not belong in a mandatory regional policy layer by default.

This is the logic of Minimum Initial Specification, Localized Future Decision, and Voluntary Adoption. It is also the logic of the Bill of Rights of Uniqueness Coordination. The registry may record. It may coordinate. It may protect uniqueness. It may not rule.

The test is severe. What breaks in the running Internet if this rule is not centralized? If the answer is not uniqueness, interoperability, registry accuracy, security integrity or operational continuity, the rule is not coordination. It is power.

Final judgment

The multi-stakeholder model was useful as transitional language. It helped resist crude state capture, broadened access and preserved some technical autonomy. But transitional language became constitutional illusion when it claimed to represent everyone while representing nobody as principal. That is the failure now visible in the numbers.

Fifteen ALAC members become the organizational home of billions of individual users. Forty-seven people become an average ICANN85 in-person session. One hundred and twenty-seven APNIC member organizations become a regional policy event. A RIPE GM turnout of 5.3% becomes an internal governance fact, not a public mandate.

WSIS+20 did not cure this defect. It made the forum permanent while leaving the execution layer unreformed. Governments received a standing forum, but not portability, failover, liability symmetry or control over private chokepoints that affect national infrastructure.

None of this is wrong if treated as advice, procedure or administration. All of it becomes absurd if treated as mandate. The Internet should stop asking only who was in the room. It should ask who can bind the party bearing the loss.

Let stakeholders speak, engineers define technical invariants, operators decide what they operate, states handle public law, markets price scarce assets, courts resolve disputes and registries record. But do not let a process call itself the people. Do not let a policy room become a legislature. Do not let a database become a throne.

That was the multi-stakeholder model’s lie. Not that stakeholders did not exist. They did. The lie was that their existence created one administrator’s mandate.

Publication source notes