Summary

  • The Bylaws Review Committee’s 2026 draft retains, rather than creates, Article 11.5. The same next-meeting endorsement checkpoint, prospective stop after non-endorsement and preservation of earlier actions appears in AFRINIC’s operative 2020 bylaws, while the draft’s own explanatory note says Article 11 has no proposed substantive change beyond editorial and consequential amendments.
  • Saving completed registry acts can be necessary when innocent parties have relied on an allocation or transfer and reversal would create duplicate claims or disrupt running networks. It does not follow that every denial, lock, suspension, refusal, committee appointment or other reversible or adverse act deserves permanent effect.
  • AFRINIC is only a private bookkeeper and technical coordinator. It has no sovereign, regulatory, police, punitive or confiscatory authority. Neither a Public Policy Meeting endorsement nor the words “shall remain valid” can manufacture such authority, extinguish a legal challenge, or turn registration into punishment.
  • A defensible Article 11.5 design needs an exact action ledger, advance notice, a reliance and reversibility classification, an automatic stop instruction, independent review, corrective remedies and an express reservation of questions about authority, legality, compensation and liability.

The moment after the meeting says no

Imagine the Public Policy Meeting has reached the decision contemplated by Article 11.5 and the answer is non-endorsement. The emergency policy can no longer be enforced or implemented from that point forward. The instruction sounds clean: stop. Yet the registry is not returned to the position it occupied when the policy began. A resource allocation may already be in a database and in routing. A transfer may already have a recipient, counterparties and customers. A request may have been refused. A registry status may have been locked. A committee may already have acted. The rule loses its future, but the facts it produced keep theirs.

That is the narrow subject of Article 11.5 and of this analysis. The provision requires a policy adopted under the preceding clause to be put to the Internet Community for endorsement at the next Public Policy Meeting. If endorsement is not obtained, the policy is not to be enforced or implemented thereafter. Actions taken before that point, however, “shall remain valid.” The endorsement checkpoint, the prospective stop and the saved-actions limb form one mechanism. They cannot be understood honestly if the last limb is treated as housekeeping.

The mechanism creates a temporal advantage. The institution acting before the checkpoint gets a window in which implementation can alter operational and economic positions. The constituency acting at the checkpoint can close that window, but its rejection does not, on the face of the clause, reopen every decision made within it. A “no” can therefore prevent tomorrow’s application while leaving yesterday’s allocation, denial, registry modification or procedural act in place. Depending on the act, that may be the only safe result—or it may make review too late to be meaningful.

This is not an argument for indiscriminate rollback. Number-resource coordination depends on uniqueness, accurate records and stable operational expectations. Erasing a completed allocation after a recipient has announced the addresses, connected customers and entered contracts can create a second crisis rather than cure the first. But the need to protect a running network does not justify saving everything an institution did during the same period. The constitutional craft lies in separating an innocent operator’s continuity from the institution’s immunity.

A retained provision, not a new 2026 invention

The first discipline is factual. The document published in 2026 is a Proposed Amended Constitution for African Network Information Centre (AfriNIC) Ltd. It is a 53-page proposal. Article 11 appears on PDF pages 20 to 22, and the corresponding explanatory note appears on pages 47 and 48. The official server recorded the PDF as last modified on 4 August 2026. As of the evidence cutoff on 10 August, the available public record did not establish member adoption, filing, an effective date or replacement of the 2020 constitution.

The Bylaws Review Committee’s draft did not invent Article 11.5. Its explanatory note says that no substantive changes are proposed to Article 11 beyond editorial and consequential amendments. AFRINIC’s operative 2020 bylaws page already contains the same architecture: submission at the next Public Policy Meeting, prospective non-enforcement following non-endorsement, and continued validity for actions already taken. The draft changes presentation and terminology, but it carries forward the substantive asymmetry.

That distinction matters because a false “new safeguard” story would obscure the real governance choice. The current question is why a rule that has already existed is being retained without a more precise account of the actions it saves or the remedy it leaves available. Retention can be an affirmative constitutional decision even when the words are old. A drafting committee that labels changes editorial still gives readers a reason to examine the unchanged machinery, especially when the machinery allocates the consequences of an unsuccessful endorsement.

The historical record shows that this machinery has attracted attention before. In 2020, a Governance Committee change table treated the saved-actions language as existing text. It discussed adding non-response as another condition that would stop the policy and considered deleting the preservation of prior actions. Its rationale warned that a policy operating until rejection could permit arbitrary implementation and potential abuse of power. That warning proves that the design risk was identified in an institutional record. It does not prove that an abuse occurred, that a proposal was adopted, or that a court found any conduct unlawful.

The operative 2020 page and the 2026 draft both retain the saving text. The continuity of the words does not settle the wisdom of the result. It tells us, instead, that the design has survived a prior debate and now deserves a more exact test: which reliance deserves protection, what remains reviewable, and who bears loss when the checkpoint arrives after implementation.

What the clause does—and what it leaves undefined

Article 11.5 supplies a sequence but not an operating manual. First comes implementation under an emergency policy. Next comes submission for endorsement at the next Public Policy Meeting. Non-endorsement then ends enforcement and implementation prospectively. Prior actions remain valid. The clause does not itself define “action,” and that omission carries much of the risk.

An action could be a completed allocation that has entered routing tables and customer contracts. It could be a transfer recorded for a recipient who paid value and reorganised a network around it. It could also be a refusal, a compliance label, a suspension, a registry lock, an interruption involving RPKI or reverse DNS, a decision not to recognise a successor, a documentation demand, a queue classification, or the appointment of an internal committee. These acts do not have the same effect, the same reversibility, the same reliance profile or the same claim to continuity.

The clause does not state a threshold of reliance. It does not distinguish an innocent third party from a participant who knew a decision was contested. It does not separate an act needed to prevent duplicate claims from a reversible administrative status. It does not specify notice to an affected party, a compensation principle or a correction mechanism. It does not create an appeal for someone harmed before the meeting. The phrase that saves actions therefore reaches further, linguistically, than the institutional safeguards surrounding it.

The endorsement side is also underspecified within Article 11.5 itself. The clause does not state the denominator, voting method, consensus test, treatment of abstentions, evidentiary record or decision-maker for a disputed result. Nor does it set a deadline for the possibility that the next Public Policy Meeting is delayed or not held. These are not abstract drafting curiosities. The duration of the pre-endorsement window determines how many facts can be created before scrutiny, while the method of closing that window determines whether the stop can be triggered clearly and promptly.

A 2020 AFRINIC policy impact assessment supplied a revealing scenario. It used terms of reference for an Appeal Committee to illustrate how an emergency Board policy might operate, later fail endorsement, cease for the future, yet leave earlier actions valid and create a subsequent procedural gap. It was a staff scenario, not proof that the emergency-policy power was actually invoked in that instance. Its value is analytical: it shows that preserving acts while extinguishing their governing rule can leave an institution with outcomes that survive but no continuing process adequate to address their consequences.

The available sources do not conclusively identify an actual case in which the Board invoked the relevant emergency-policy route, named a specific policy, took actions under it, obtained or failed endorsement and then applied Article 11.5’s savings limb. Actual invocation history remains open. It would be equally wrong to invent a completed case and to declare that no invocation has ever occurred. The evidence supports examination of the text, historical debate and documented scenario—not a claim about a real endorsement result that has not been established.

Nor is this the place to decide whether any emergency was urgent, necessary or properly identified, or whether the policy-making power in Article 11.4 was properly used. Those are separate questions. Article 11.5 begins after a policy has entered its track. Its concern is the checkpoint and the allocation of consequences when endorsement does not follow.

Five classes hidden inside “any action”

A serious savings rule must look through the generic noun and identify the actual thing being preserved. Five classes provide a workable starting point.

The first is an irreversible or high-reliance uniqueness act. Consider a completed allocation that is already routed and supports customers, or a registry record whose deletion could produce duplicate entitlement or an immediate outage. Preserving operational effect here protects people who did not control the institution’s process. It keeps the registry accurate about the network that now exists. Independent review should remain available for authority, harm and liability, but retroactive erasure may be a reckless remedy. The objective is continuity, not vindication of the decision-maker.

The second is a reversible administrative act. A pending workflow status, a request for documentation, a future eligibility classification or a non-final place in a queue normally creates less legitimate reliance. Once the policy is not endorsed, such an act can be frozen, examined and corrected. Treating it as permanently valid simply because staff clicked a field before the meeting would elevate internal timing over substantive justification. Reversibility is a reason to repair, not an invitation to destabilise networks.

The third is an adverse or coercive registry act. A transfer refusal, resource lock, revocation step, service suspension, interruption to RPKI or reverse DNS, adverse compliance label, or refusal to recognise a successor can impose costs before review occurs. Those costs may include delayed transactions, lost customers, legal expenses, renumbering risk and operational uncertainty. Such acts require notice, reasons, continuity protection, independent review and an effective corrective remedy. Article 11.5 cannot convert technical custody of registry records into punitive authority.

The fourth is a third-party transaction act. A transfer or allocation may have induced a recipient, lender, customer or counterparty to change position. Protecting innocent reliance and uniqueness can be necessary even if the underlying decision remains disputed. Yet preservation of operational effect should not foreclose questions about mistake, fraud, conflicts, authority, compensation or correction. A safe harbour for an innocent recipient is not a pardon for the institution. It is a technique for keeping one dispute from multiplying into several network failures.

The fifth is an internal process act: an appointment, terms of reference, a hearing step or a procedural determination. Completed fair-hearing work and evidentiary records may remain useful after a policy stops. But a rejected policy should not create an indefinite mandate merely because a committee met before the checkpoint. Internal process is generally more capable of correction than an address block already embedded in production networks. The presumption should therefore be preservation of evidence, not automatic preservation of institutional office or authority.

This taxonomy is not a judicial code. It is a governance discipline built around consequence. It asks whether reversal threatens uniqueness, whether an innocent person relied, whether the act is technically reversible, whether the act burdened rather than benefited an operator, and whether continued effect would enlarge AFRINIC’s role beyond coordination. Those questions make the phrase “shall remain valid” precise enough to serve networks without becoming a blank cheque.

The distinction also improves decision quality before the meeting. If implementers know each act must be logged and classified, they cannot hide a punitive lock and an operational allocation in the same undifferentiated category. If an adverse act will be stayed unless delay itself threatens uniqueness or security, the institution has less incentive to front-load harm. If high-reliance acts require written reasons identifying the reliance at stake, preservation becomes a demonstrable necessity rather than an institutional reflex.

The strongest case for a prospective-only remedy

The strongest contrary case deserves to be stated without caricature. A uniqueness registry cannot treat every internal procedural defect as a command to erase the operational past. Number resources are not entries in an isolated spreadsheet. Once a completed allocation or transfer is reflected in routing, security systems, contracts, financing and customer services, retroactive invalidation can create competing claims and inaccurate records. It can strand an innocent recipient, interrupt customers and encourage strategic challenges to every decision whenever governance is disputed.

On that view, Article 11.5’s asymmetry is a prudent stabiliser. The Public Policy Meeting retains the ability to stop an unendorsed rule. The network retains the decisions on which innocent parties have already relied. The institution avoids pretending that a late procedural decision can make elapsed time disappear. Prospective correction is familiar precisely because retroactivity can be more unjust than the error it seeks to cure.

This case is especially strong where uniqueness itself is involved. Two parties cannot safely receive the same exclusive registration simply because a governance review later changed direction. A recipient who has deployed an address block is not a prop in an institutional argument. Customers and counterparties may have had neither notice nor influence. Running networks deserve priority over procedural theatre. Accurate records should describe operational reality, not perform a symbolic rewind that creates collisions.

The contrary case also warns against an overbroad remedy culture. If every saved act is treated as suspect merely because the policy later lacked endorsement, registry decisions become provisional in a commercially destructive way. Transaction costs rise. Parties demand discounts or guarantees against constitutional uncertainty. Liquidity suffers. Better-connected actors may be able to price the risk while smaller operators cannot. A stable prospective stop can contain the dispute and protect people outside it.

All of that is true, and none of it closes the case. The argument establishes a reason to preserve reliance-heavy and continuity-critical acts. It does not establish a reason to preserve every reversible status, every denial, every coercive restriction or every institutional appointment. The word “any” collapses the difference between protecting an innocent recipient and immunising the actor that created the problem. A tailored reliance safe harbour answers the continuity case more accurately than blanket validity.

The first-mover advantage inside the savings rule

The central incentive problem is simple. If the policy can operate until the checkpoint, and everything done before non-endorsement remains valid, the actor controlling implementation has an option: act quickly and create durable facts. Those affected must wait for a meeting to stop future conduct, but they may have to pursue separate review, negotiation or litigation to undo what has already happened. The formal checkpoint exists, yet its practical value diminishes with each completed action.

That is path dependence. A decision enters operational systems. A counterparty relies. A status changes. A customer is lost. Costs accumulate around the new position. By the time the constituency says no, reversal is more expensive and the institution can point to the very reliance its own speed helped create. The savings rule then rewards the first mover—not necessarily because the original act was justified, but because time has made it harder to correct.

This asymmetry has economic consequences under IPv4 scarcity. A preserved allocation can become embedded in financing, transactions and network design. A preserved denial or lock can delay a transfer, reduce an operator’s bargaining position and impose opportunity costs that a later prospective stop does not repay. An actor with privileged access to the institutional process can anticipate timing better than an outsider. Uncertainty therefore taxes ordinary operators while rewarding those able to navigate, influence or absorb governance risk.

The effect is not limited to resource recipients. Staff who implement a policy face ambiguity about which systems must stop and when. Queued cases may straddle the decision time. Cached classifications may continue to influence later work. Security and directory services may not update simultaneously. Without an exact stop timestamp and a technical rollback plan, prospective cessation can be nominal while pieces of the rejected rule continue through automated or inherited processes.

Blanket saving also weakens the incentive to design proportionate interim measures. If every completed act is guaranteed survival, there is less institutional reason to prefer temporary, reversible and minimally harmful steps. A narrow rule would reverse that incentive. It would tell the purported/current Board and implementers that continuity-critical outcomes can be protected, but adverse choices, unresolved queues and institutional mandates will face review. Speed would no longer be a substitute for legitimacy.

Bookkeeping cannot become punishment

The constitutional boundary begins with the nature of the institution. AFRINIC is only a private bookkeeper and technical coordinator. It may maintain accurate records, coordinate uniqueness and protect the continuity of technical registration. It has no sovereign, regulatory, police, punitive or confiscatory authority. Number-resource administration does not make it a government, and community process does not make meeting participants a legislature over absent operators or assets.

This truth determines how “valid” must be read. The word can preserve an act’s operational effect within the registry while continuity and lawful review are arranged. It cannot manufacture authority that AFRINIC never possessed. It cannot convert an ultra vires, unlawful, fraudulent, conflicted or court-restrained act into a lawful one by constitutional declaration. It cannot extinguish a court’s power, a contractual claim, a claim for compensation or a challenge to the decision-maker’s authority.

Consensus is evidence, not sovereignty. Endorsement may demonstrate technical support and satisfy an internal governance condition. Non-endorsement may trigger the clause’s prospective stop. Neither result creates ownership of number resources or a public-law mandate to decide an operator’s economic fate. When registry control is used to punish a business model, block capital, or impose an institutional orthodoxy unrelated to accurate coordination, bookkeeping has become capital control.

The same limit applies to a saved act. A resource lock does not become non-punitive because it happened one minute before the checkpoint. A refusal does not become insulated because the policy later stopped. Registration may not be turned into punishment through timing. The appropriate response protects the integrity of records and running services while keeping adverse action open to reasons, review and correction.

Power and liability must remain coupled. If AFRINIC preserves the consequences of a discretionary act while externalising all loss to an operator, counterparty or customer, it keeps power and discards responsibility. A registry that exercises consequential discretion must accept proportional accountability. If it will not accept review, correction and possible compensation, it must surrender the discretionary power rather than use technical dependency as leverage.

This is why operational continuity and institutional immunity must be separated in the drafting. Continuity asks: what must remain in place today so that unique registrations, security functions and customers are not harmed? Immunity asks: who can still challenge the decision, obtain reasons, secure correction or recover loss? The first may justify temporary preservation. It never requires the second to disappear.

The purported/current Board and the danger of retrospective cure

Authority matters more when a savings clause allows decisions to outlive the policy that produced them. The Bylaws Review Committee’s terms say the purported/current Board established the committee under Article 15.3(x), appointed its chair, receives its reports, reviews its proposals and prepares them for member consideration. That official page proves the process AFRINIC claims. It cannot, by itself, prove that the appointing Board’s authority is uncontested.

AFRINIC stated on 12 March 2026 that the people it presents as its Board had resumed duty while collaborating with the court-appointed receiver pending his formal discharge. AFRINIC’s court-case list, last modified on 15 June in the checked record, listed the October 2025 application for the receiver’s discharge as ongoing. It also listed a March 2026 challenge concerning Board ratification of the transfer policy as ongoing. These entries establish recorded proceedings and AFRINIC’s acknowledged transition account. They do not decide the merits or supply a final court result.

NRS’s formal member-action statement of 24 June described the current directors as a Purported Board, challenged their meeting authority and rejected retrospective cure of disputed acts through member approval. Its immediate subject was the 2026 Annual General Members’ Meeting and legal expenditure, not Article 11.5. The statement is nevertheless first-class authority-risk context because it puts the no-retrospective-cure issue squarely into the present governance record. NRS advocates, researches, convenes and represents explicitly authorised members; it does not run AFRINIC’s registry or decide endorsement.

The correct formulation is therefore cautious but not evasive. Where present authority is material, the body is the purported/current Board. Its authority is contested in a formal member action and exists alongside official records acknowledging an unfinished receiver transition and ongoing proceedings. The available evidence contains no final holding that resolves those questions. It would be wrong to call the Board conclusively unlawful, and equally wrong to use the institution’s own process documents as self-proving authority.

Article 11.5 must not become a route for retrospective cure. A member vote or Public Policy Meeting endorsement can have whatever internal effect valid corporate and policy procedures lawfully assign to it. It cannot rewrite history so that an act by an unauthorised body was always authorised, nor can non-endorsement plus saved validity decide the matter against an affected party. Questions about corporate authority and legal effect belong to competent adjudication or a genuinely independent agreed reviewer, not to a savings clause interpreted by its beneficiary.

The 19 April consultation notice described an indicative route from initial comments to draft publication, public comment, final analysis, a final report, notice and a planned September Special General Members’ Meeting. That sequence reinforces the distinction between consultation, proposal, member consideration, filing, effective operation and later use. A draft is not an adopted constitution. The possibility of later member consideration does not permit anyone to report the 2026 text as already in force or to infer the outcome of a vote not established by the record.

A narrow design that preserves networks and permits correction

The sound response is neither blanket erasure nor blanket validation. It is differentiated preservation. An act that created substantial innocent third-party reliance, or whose immediate reversal would threaten uniqueness or a running network, should keep operational effect while authority and harm are independently reviewed. A reversible administrative act should be frozen and corrected. An adverse or coercive act should receive heightened scrutiny, notice and a remedy. An internal mandate should not become indefinite merely because its first meeting occurred before rejection.

Every act taken in the pre-endorsement interval should enter a public register. The entry should identify the policy, the Board resolution, the adoption time, the implementing system, the affected class and a named institutional owner. It should state whether the act is high-reliance, reversible, adverse, transactional or internal. It should record what reliance existed at the moment of decision rather than allow reliance to be manufactured later as a defence.

Affected parties should receive notice that the decision operates under an emergency policy subject to Article 11.5 review. The notice should identify the expected checkpoint, the possibility of a prospective stop and an objection channel. For an adverse act, reasons should be available promptly enough for review to matter before economic loss hardens into a permanent fact.

The clause also needs an operational stop, not just a legal sentence. The responsible body should define the exact timestamp at which future enforcement ceases after non-endorsement. It should identify queued cases, cached classifications and each technical system that must be changed. A named owner should certify the stop. Otherwise, the rejected rule may linger in software, workflows and staff assumptions even though the formal policy has ended.

Independent review should be able to preserve continuity, demand reasons, order correction, provide interim protection and refer legal questions to a competent court where necessary. The reviewer must be independent of both the purported/current Board and those who implemented the act. Review should distinguish operational preservation from legal validation: an allocation can remain active to prevent disruption while liability or compensation is considered.

Finally, the constitutional text or its implementing framework should reserve challenges expressly. Continued operational effect must not waive questions of Board authority, applicable law, contractual rights, fraud, mistake, conflict, damages, compensation or a court order. Such a reservation would not destabilise the registry. It would tell every participant exactly what preservation means—and what it does not.

What can be said now

The evidence is sufficient to establish the design. Article 11.5 appears in the 2026 proposal; it requires the next-Public-Policy-Meeting endorsement checkpoint, stops the policy prospectively after non-endorsement and preserves earlier actions. The operative 2020 page has the same substantive mechanism. The draft’s explanatory note says Article 11 has no proposed substantive change beyond editorial and consequential amendments. A 2020 change table records prior debate, and a staff assessment supplies a concrete hypothetical about saved acts outliving a rejected policy.

The evidence is not sufficient to establish an adopted replacement constitution, a completed Article 11.5 case, a defined endorsement method, a public ledger of interim actions, a final resolution of current Board authority or the legal reach of the phrase “shall remain valid.” An authenticated redline, Article 11.5 consultation responses, vote rules, delayed-meeting rules, actual invocation records, technical stop procedures, remedy terms and final court determinations remain missing from the available public record.

That uncertainty does not make analysis premature. It makes precision mandatory. The retained clause can be evaluated without pretending it has already been used or that pending authority disputes have been adjudicated. Its legitimate purpose is narrow: protect uniqueness, accurate records, innocent reliance and running networks. Its danger is equally clear: if every pre-review act survives without classification or remedy, the checkpoint can stop the words while leaving the exercise of power intact.

The right constitutional principle is therefore not that the past is untouchable. It is that continuity-critical facts should be preserved in the least harmful form, while reversible and adverse decisions remain correctable and their makers remain accountable. AFRINIC’s technical role is important precisely because it is bounded. A private bookkeeper can coordinate uniqueness. It cannot rule.