Summary

  • AFRINIC's public record for the September 2025 replacement election reported 581 voters, 548 completed biometric registration and 484 votes cast. Those figures expose two distinct gaps: 33 people did not complete the identity stage, while 64 completed it but did not cast a ballot. The published totals do not show how many in either group chose not to proceed, encountered a technical problem, failed a rule or sought review.
  • The June election had a different access design, combining electronic voting with an in-person polling station and documentary routes for authorised representatives, proxies and powers of attorney. The receiver later annulled the results after reporting suspicions of irregularities, while expressly saying an investigation had not reached final conclusions. Annulment resolved the election as a whole but did not publish a voter-by-voter account of admission, refusal, complaint or remedy.
  • Identity verification is only one gate. A person can be the human shown on an identity document and still lack corporate authority; an authorised representative can still be linked to an ineligible member; and an eligible, properly designated voter can still fail to receive or use a functioning ballot credential.
  • A credible exclusion record must begin with the full eligible-member denominator and preserve a reasoned path through notice, designation, document review, identity verification, platform activation, attempted access, ballot acceptance, challenge and final disposition. It must distinguish voluntary non-participation from administrative exclusion.
  • Privacy does not require institutional blindness. AFRINIC could publish aggregate reason counts and an independently certified reconciliation while giving each affected member a private decision record, evidence route, rapid appeal and final answer before voting closes.

The election is also the queue outside it

Election reports usually begin with votes cast. They show totals, percentages, successful candidates and perhaps turnout. That is the visible election. The less visible election consists of every organisation and person that approached the franchise but did not reach a counted ballot.

Some never received notice. Some were not on the eligibility list. Some were told that a fee, membership date or company record disqualified them. Some submitted a voter designation but could not prove the signatory's authority. Some reached an identity check that could not match their document or face. Some activated an account but did not receive a credential. Some encountered a device, network or support failure. Some simply changed their mind. These are different events. A single turnout percentage erases the distinctions.

That erasure mattered in AFRINIC's 2025 board restoration because the election was not routine. The organisation had spent years without an ordinary quorate board. A court-appointed receiver was responsible for organising the route back to member governance. Every decision about who could participate therefore helped determine who would acquire authority over the company after the exceptional period.

The title of this article describes a class of people, not a proven count. The reviewed public materials do not supply a complete list of individuals who physically arrived at the June polling station or digitally reached the September platform and were refused a ballot. That absence is the problem under examination. If an election publishes accepted voters without preserving the people stopped at each gate, it cannot distinguish a clean franchise from an orderly-looking result produced after invisible exclusion.

The correct question is not whether administrators intended to exclude anyone. It is whether the institution can account for every transition from potential member voter to counted ballot. Intent cannot repair missing records. A fair decision can look arbitrary when reasons disappear; an unfair decision can look routine when only the successful entries remain.

The denominator must precede the allegation

A responsible exclusion analysis starts by refusing dramatic shortcuts. It does not infer that every non-voter was blocked. It does not call every failed document fraudulent. It does not treat every unanswered email as proof of suppression. It constructs the denominator first.

For AFRINIC, at least six populations must be counted separately. The broadest is the population of organisations recorded as members or resource holders. The next is the population assessed under the election's eligibility rule. The third is the eligible organisations that received effective notice. The fourth is the organisations that submitted a voter designation or other authority document. The fifth is the human voters accepted and activated by the voting system. The sixth is the voters whose ballots were accepted and included in the result.

Each transition has both additions and losses. A legal-entity correction may add a member to the eligibility review. A cutoff rule may exclude an otherwise current member. A late fee may restore good standing if the rule permits cure. A designation may fail because the signatory's office cannot be confirmed. An identity check may reject an expired document. A platform account may be activated but never used. A submitted ballot may be quarantined because a duplicate credential is detected.

Without these layers, numbers become misleading. Suppose an election reports 484 votes from 581 registered voters. That supports a turnout calculation among registered voters. It does not reveal how many eligible organisations never entered registration, how many sought entry and were refused, or how many activated voters attempted to cast a ballot but could not. It certainly does not establish why 97 registered voters did not produce a vote.

The discipline is symmetrical. A critic should not inflate the excluded population by treating every non-vote as denial. An administrator should not shrink it by reporting only formal rejection letters. The correct data include voluntary withdrawal, no response, late filing, unresolved evidence, rejection, technical failure, successful appeal and every other final state. Only then can exclusion be measured rather than narrated.

June used several paths to the same ballot

The June 2025 arrangements combined electronic voting with in-person voting scheduled for 23 June. AFRINIC's published materials described documentary routes for representatives and proxies. A communiqué issued on 18 June said that an appointed proxy attending the polling station needed a signed authority letter, the prescribed proxy material or a notarised copy, and current identification. A person unable to produce the required documents would not be eligible to vote in person by paper ballot.

Those requirements had an intelligible purpose. A company does not enter a polling station; a human acts for it. Election administrators needed to establish both the person's identity and the chain by which the member organisation authorised that person. Paper ballots also required controls against duplicate voting across electronic and in-person channels.

The design nevertheless created several possible refusal points on election day. Was the organisation on the relevant member or voter list? Was the attending person the named representative? Did the authority letter come from an acceptable corporate officer? Did the proxy form satisfy the constitution and published guidance? Was a power of attorney treated as a separate route, and if so under what scope? Was the identity document current and consistent? Had an electronic vote already been recorded? Had one person reached a limit on proxies? Could a defect be cured while the polling station remained open?

Public discussion later concentrated on powers of attorney and the possibility that one person could carry authority for multiple organisations. ICANN's 25 June letter asked detailed questions about how those documents had been accepted and how in-person ballots had been controlled. Stakeholder statements raised related concerns. These materials establish that questions were asked; they do not adjudicate any individual document or ballot.

The receiver's 15 July notice is equally bounded. It said suspicions of irregularities had been raised, particularly concerning powers of attorney, that complaints had gone to relevant authorities and that police were investigating. It also said final conclusions were absent and the receiver could not formally report the extent of irregularities. The decision was to annul the results and hold a new election.

This sequence is important for exclusion analysis. It confirms that documentary authority was a decisive control surface. It does not tell the public how many people presented themselves, how many were accepted, how many were refused, which reasons applied, whether a supervisor reversed any refusal, or whether rejected people submitted complaints. The election was handled globally through annulment, while the individual access record remained publicly unresolved.

Arrival is evidence, not entitlement

Showing up at a polling place or opening an election application does not by itself create a right to vote. An election must enforce eligibility rules. The challenge is to make those rules prior, intelligible, proportionate and reviewable.

Four propositions can all be true. First, an organisation may be a genuine AFRINIC customer or resource holder. Second, it may not satisfy the particular membership class or cutoff used for a board election. Third, a human may genuinely work for that organisation. Fourth, that human may still lack authority to cast its corporate vote. Administrative fairness requires that each proposition be decided by the appropriate evidence rather than by intuition at the final gate.

The June polling station was a poor place to discover an upstream problem. By then a person may have travelled, assembled documents and relied on prior communications. A refusal based on a member-list omission or disputed corporate record could not always be cured in minutes. The official at the desk might be competent to inspect identification but not to decide a company-law dispute. Delay would also affect other voters and the closing time.

This is why pre-clearance should do most of the work. Every proposed voter should receive a status before the election: cleared, conditionally cleared with a specified missing item, rejected with reasons, or under expedited review. The institution should state exactly what must be brought on the day and which defects can be cured there. The polling official then verifies a known record rather than inventing an eligibility decision under pressure.

There must still be an exception channel. Documents can be lost, names can differ after marriage, a corporate officer can change, or an administrator can make an error. A provisional or challenged ballot can preserve the vote without immediately accepting it into the count. The ballot remains secret and sealed while an independent reviewer decides eligibility. If the challenge succeeds, it is included; if it fails, the rejection and reason are recorded without exposing vote choice.

Arrival is therefore probative. It shows attempted participation and may reveal defects in notice or pre-clearance. It is not conclusive entitlement. A mature election records both the attempt and the lawful reason for the final result.

Identity technology answers a narrow question

The September replacement election removed proxies and powers of attorney, required one designated voter per eligible Resource Member and moved voting online. The published designation material described executive sign-off, corporate contact details, identity documents, authenticity checks and facial biometric comparison through a third-party provider.

This reduced some June risks. A single human could no longer arrive carrying a chain of proxies. The member organisation had to name one voter. The voting provider could compare the person using the account with a government-issued identity document. A digital record could preserve timestamps for registration, verification and voting.

But stronger identity proof can create false confidence if it is treated as proof of every upstream fact. Facial comparison can indicate that the person presenting the credential resembles the person on the document. It cannot prove that a chief executive validly designated that person. It cannot prove that the signatory was actually chief executive. It cannot prove that the organisation was the legal member, that fees were correctly classified or that the election cutoff was lawfully applied.

Identity checks also produce their own error classes. A document may be unsupported, expired, damaged or difficult for automated systems to read. Names may use different scripts or ordering. Image quality, lighting, camera access, device compatibility, disability, age-related appearance change and weak connectivity may affect completion. A fraud-resistant design must not convert a vendor's uncertain match into an unreviewable loss of corporate voting rights.

The response should be graduated. A strong match can pass automatically. An inconclusive result should move to trained human review using minimum necessary data. A suspected document alteration should trigger preservation and a reasoned decision, not a public accusation. A voter who cannot use facial comparison should have an equivalent secure route. Every final rejection should identify whether the problem concerned document validity, person match, corporate authority, duplication or another specific rule.

Privacy matters throughout. Biometric material is unusually sensitive and should not become a permanent membership credential. The election provider should collect only what the defined verification needs, protect it, limit access, set a deletion period and retain only the evidence necessary to prove the result and resolve a timely challenge. An audit does not require publishing faces or identity documents. It requires proving that a consistent decision method operated.

The September numbers reveal where questions begin

AFRINIC's election statistics reported 581 total voters, 548 completed biometric registration and 484 votes cast as of 12 September 2025. These are unusually useful numbers because they show more than the final tally. They also demonstrate the limits of aggregate reporting.

The first subtraction is direct: 33 registered voters did not complete biometric registration. The statistics page does not, by that number alone, establish why. Some may never have attempted it. Some may have started late. Some may have encountered a technical or document problem. Some may have decided not to vote. Some may have sought help. Calling all 33 excluded would be unsupported; calling all 33 voluntary abstainers would be equally unsupported.

The second subtraction is 64: 548 people completed biometric registration, but only 484 votes were cast. Again, the difference is not self-interpreting. A verified voter may consciously abstain. A voter may forget the deadline, disagree with all candidates, lose access to a device, fail to receive a final credential, encounter a platform error or believe a ballot was submitted when it was not. Some may have contacted support, while others may have remained silent.

The two gaps should not be merged into a single group of 97 non-voters. The 33 stopped before or at completion of identity verification. The 64 passed that stage but did not produce a counted vote. The remedy for a document-reading problem is different from the remedy for a missing voting credential or a deliberate abstention.

There is also an upstream denominator not shown by these three figures: eligible organisations that never became one of the 581 registered voters. The final voters register described organisations and designated voters, but the public statistics do not provide a complete reconciliation against every member assessed for eligibility. Without that number, the election can calculate participation among registered voters but not full participation among all organisations entitled to seek a vote.

The figures are not evidence that the result was invalid. They are a map of unanswered categories. Their value would rise sharply if AFRINIC published aggregate reason and support data: how many never attempted identity verification, how many failed automatically, how many went to human review, how many appealed, how many were restored, how many credentials were issued, how many login or ballot incidents were opened, and how many remained unresolved when voting closed.

Reasons must be operational, not moral

An exclusion ledger should use specific reason codes tied to rules and evidence. It should not use broad labels such as suspicious, non-compliant or invalid when a narrower description is available. Broad labels create stigma and conceal whether the defect was curable.

At the member layer, possible reasons include not in the relevant membership class, admitted after the governing cutoff, terminated status, unresolved legal-entity succession, or fees unpaid by a defined date. Each requires a reference and the rule applied. A payment dispute should not be described as identity failure; a disputed merger should not be described as non-cooperation.

At the authority layer, reasons include missing designation, unacceptable signatory, inability to confirm the signatory's office, conflicting designations, revoked authority or a document that does not authorise the relevant act. The administrator should distinguish a missing page from a substantive lack of corporate power. One may be cured quickly; the other may require independent legal evidence.

At the human identity layer, reasons include unsupported document type, expired document, inconsistent name, unreadable image, inconclusive facial comparison, duplicate identity, or suspected alteration. An automated mismatch should be recorded as inconclusive until reviewed, not as fraud.

At the platform layer, reasons include account not activated, credential not delivered, authentication failure, unsupported device, connectivity interruption, duplicate account, ballot not completed, ballot rejected before submission or ballot accepted with receipt. Technical-support contacts should be linked to these states without exposing ballot choice.

At the timing layer, the record must distinguish late first application, timely application with late institutional review, timely application missing evidence, late confirmation requested by administrators, and delay caused by the provider. The final-register page said four designated voters were added after late confirmations while no new applications were accepted after the deadline. That distinction may be proper, but it should appear as a reasoned exception class applied equally to comparable cases.

Specific codes improve fairness and management. They reveal whether many members misunderstood the same instruction, whether one region faced document incompatibility, whether a provider produced concentrated failures or whether institutional review rather than applicants caused delay. The data can improve the next election without accusing individuals.

An appeal must preserve the possibility of voting

An appeal that ends after the election is an explanation, not an electoral remedy. Once candidates take office, a successful excluded voter usually cannot receive the ballot opportunity that was lost. The institution may apologise or rerun the entire election, but both are costly substitutes for timely correction.

The review clock should therefore match the election clock. Eligibility and designation decisions should be issued early enough for a first challenge. Identity verification should open with enough time for human review. The platform should provide a live access channel during voting. A final emergency reviewer should remain available until the ballot closes.

Each affected member needs a private decision notice containing the rule, decisive facts, missing or rejected evidence, submission deadline, review body and expected response time. The member should not have to infer its status from absence on a public list. Silence is especially dangerous where old contact data may be the reason the member never knew to act.

Review should be independent of the original decision in contested cases. Staff may correct obvious clerical errors immediately. A separate reviewer should decide disputed membership, authority or identity conclusions. Where the receiver established the election, the receiver should not be the only final reviewer of every access decision. An election committee, external assurer or court-directed referee can provide separation, subject to the exact legal mandate.

Provisional ballots or sealed digital voting credentials can protect time. If the reviewer cannot decide before closing, the person may cast a segregated ballot whose inclusion depends on the later decision. This is suitable only where secrecy and tally integrity can be preserved and where the number of provisional ballots cannot reveal individual choices in a small contest. The rules must be published in advance.

There also needs to be a remedy ladder. A curable document defect leads to resubmission. An institutional delay leads to an extended individual deadline. A vendor failure leads to an alternative secure channel. A wrongly denied voter receives immediate access. A material group exclusion may justify extending or reopening voting. Annulment is reserved for defects whose reach cannot be isolated or whose effect makes a reliable result impossible.

June's annulment was complete; its explanation was not

The receiver chose the broadest electoral remedy after the June concerns: the results were annulled and a new election ordered within an extended mandate. This eliminated the legal and practical effect of all June ballots rather than attempting to remove only particular votes.

There may be circumstances in which that response is proportionate. If administrators cannot identify which ballots depend on defective authority, if secrecy prevents separation, if records are unreliable or if the number of disputed ballots could affect several contests, a complete rerun may be safer than speculative reconstruction. The public materials reviewed here do not provide enough evidence to decide whether those conditions existed.

The receiver's own notice stated that the extent of irregularities could not yet be reported because the investigation had not reached final conclusions. That is an important limitation. It prevents the annulment itself from becoming proof that every ballot, voter or document was tainted. It also leaves a governance question: what happened to people whose documents were rejected or challenged before voting was suspended?

Final handling should be recorded at two levels. The election-level outcome was annulment. The voter-level outcome should still classify every attempted vote: accepted before suspension, refused with reason, conditionally admitted, challenged, withdrawn, unresolved or never presented. Complaints should be linked to decisions and preserved for later review. A rerun does not make those records irrelevant. They are evidence about which control failed and whether the replacement design repaired it.

People who were properly excluded in June may still deserve clear reasons. People wrongly excluded may deserve acknowledgement even though everyone votes again. People whose authority documents were questioned should not remain under an indefinite cloud when no finding follows. The institution should close each case or state why it remains with an external authority.

Collective annulment can restore equality of voting opportunity while leaving individual procedural justice unfinished. A credible institution must do both.

The final register did not close every access question

For the September election, AFRINIC published a provisional register and then a final register. The public page said no discrepancy reports had been received regarding the provisional version. It also described four additional designated voters added after late confirmations from the nominating executives of eligible members, while saying no new applications had been accepted after the formal deadline.

Publication was a valuable control. It allowed included organisations and observers to inspect the names and basic status displayed. The additional biometric-completion field later made part of the access funnel visible. But a public list is much better at proving inclusion than exclusion. An omitted organisation may not know to inspect it, may not receive the notice, or may not understand why it is absent.

The statement that no discrepancies were reported must therefore be read narrowly. It reports what AFRINIC said it received through the relevant channel. It does not prove that every eligible member saw the register, understood the challenge route and had time to assemble evidence. To evaluate the correction opportunity, the public needs delivery, acknowledgement and support denominators.

The four additions also show why finality needs version history. A final register can change for legitimate reasons, but the rule allowing change must be explicit. The record should show that each underlying application was timely, what confirmation was missing, when it was requested, when it arrived, who approved the addition and whether every similarly situated applicant received equivalent treatment.

Once voting begins, the register should be frozen except for predefined mandatory events. The voting provider should record the exact version received. A later removal, credential suspension or correction should generate an immutable exception and independent approval. The final report should reconcile that version to the 581 registrations, 548 biometric completions and 484 votes.

Final does not mean beyond correction. It means that every permitted correction follows a visible rule and leaves the prior state intact.

Privacy and the exclusion denominator can coexist

Election administrators often resist detailed exclusion reporting because membership, payment, identity and biometric facts are sensitive. The concern is valid. Publishing an organisation's arrears, an individual's failed identity match or the contents of an authority document could create commercial, security and data-protection harm.

The answer is layered disclosure. Each affected organisation receives its own detailed record. The independent assurer receives protected evidence under strict access and retention controls. Members and the public receive aggregated reason counts and the reconciliation needed to assess completeness. No layer needs vote choice.

Aggregate publication could show, for example, the number of organisations assessed, eligible, notified, designated, rejected by reason, pending at cutoff, restored on review, activated, verified and counted. Small cells can be combined or delayed where disclosure could identify a person. Regions should not be published where the count would invite political inference or expose an individual member.

The identity provider should publish method-level assurance: supported document classes, manual-review availability, accessibility route, retention period, incident count and completion statistics. It should not publish biometric templates. AFRINIC should disclose who makes a final eligibility decision and who merely supplies a technical signal.

An independent certificate can confirm that the final voter register reconciled with references, that changes after freeze were authorised, that every accepted ballot mapped to one eligible voter, that no voter cast more than allowed and that every rejection had a final state. The certificate can describe exceptions without naming the affected parties.

Privacy protects people from unnecessary exposure. It should not protect an institution from having to know how many people its own rules stopped.

Build the exclusion record from events, not a spreadsheet

A flat spreadsheet is tempting because it gives administrators a final list. It is inadequate because it overwrites the history that explains how the list was produced. The stronger design is an event record linked by stable member and voter identifiers.

For each organisation, the record begins with the election eligibility assessment. It captures the rule version, cutoff, membership class, status, relevant fee conclusion, notice channels and reviewer. Every designation submission becomes a dated event. Supporting documents are referenced, not casually copied. A decision event records acceptance, rejection or a request for more evidence. A challenge event records the grounds, reviewer and response deadline.

For each human voter, separate events record identity-document submission, automated result, human review, accessibility accommodation, approval, platform activation, credential delivery and support incidents. Ballot events should reveal only issuance, attempted access, successful submission and receipt, never vote choice in the administrative record.

Every event requires a final disposition. Open cases at the start of voting are counted and either resolved, granted provisional access or expressly closed under a published rule. Open support cases at the end of voting are reviewed before certification. A case cannot disappear because a later row says approved or denied.

The public report can then reconstruct the funnel. It can say how many started at each stage, how many moved forward, how many left voluntarily, how many were rejected, how many appealed and how many succeeded. The arithmetic must balance. If 581 people were registered, every one should end in one of the recorded states. If 548 completed biometrics, the remaining 33 need aggregate states. If 484 voted, the remaining 64 need aggregate states distinct from the first 33.

This is not an elaborate addition to election administration. It is the evidence that administration happened consistently.

Exclusion data changes institutional incentives

What an institution measures affects what it treats as failure. If success means only preventing duplicate or unauthorised votes, administrators will naturally prefer strict gates. A mistaken exclusion produces little visible cost, while a mistaken inclusion can become a scandal.

If the institution must report excluded voters, unresolved challenges and restored appeals, the incentive becomes more balanced. Administrators still protect the ballot, but they also see loss of valid participation as an integrity failure. Vendor contracts can include completion, accessibility and review standards rather than only security claims.

The data also reveal distribution. A rule may look neutral while burdening companies with non-standard corporate titles, members in jurisdictions with different document formats, people with weak connectivity or organisations whose old contacts were never updated. Aggregate reason patterns can expose that effect without assuming discrimination.

Candidates and members gain a better basis for evaluating legitimacy. A high percentage of votes among activated accounts may coexist with poor notice coverage. A low turnout may reflect choice, friction or exclusion. The institution should not claim more than the denominator supports.

Most importantly, an exclusion record limits retrospective storytelling. After a contested result, each side tends to classify non-voters in the way that supports its argument. One calls them apathetic; another calls them disenfranchised. A contemporaneous event record replaces both labels with evidence.

The handover must preserve unfinished cases

An election restoring a board creates a difficult transition. The new directors benefit from the result and may inherit complaints about how it was produced. The receiver or election committee may be preparing to leave. Vendors may close accounts and delete data. That is precisely when exclusion evidence can disappear.

Certification should therefore include a sealed handover. It should preserve the final register version, eligibility decision history, designation evidence references, identity-verification outcomes, platform access logs, support cases, appeals, incident reports and retention schedule. Vote secrecy remains protected. Access to personal data remains limited.

Unresolved complaints should transfer to a body capable of independent review, not simply to the newly elected board. The board can fund and cooperate with the review, but should not decide whether its own election was properly constituted. A court-directed reviewer, election assurer or other legally valid independent body may be appropriate depending on the dispute.

Retention must also have an end. Identity documents and biometric data should not remain indefinitely merely because litigation is possible. The institution should preserve minimum proof, legal holds and cryptographic integrity while deleting unnecessary raw material under a published schedule. Complainants should know whether their matter creates a lawful hold.

The later reuse of the September 2025 register as a baseline for 2026 elections makes this handover more important. A crisis-era omission can become a future default if the old list is treated as truth. Every later confirmation period should reopen contact, authority and eligibility checks for organisations that did not participate, rather than deeming silence permanent absence.

A minimum public exclusion report

AFRINIC or any similarly situated registry could publish a useful report without exposing sensitive identities. The report should contain the following balanced counts:

  1. organisations in the relevant membership population;
  2. organisations assessed for election eligibility;
  3. eligible, ineligible, pending and disputed decisions at the first cutoff;
  4. notices sent, delivered, bounced and acknowledged;
  5. voter designations submitted, accepted, rejected, incomplete and withdrawn;
  6. identity checks started, automatically completed, manually reviewed, rejected, abandoned and restored on appeal;
  7. voting accounts activated and credentials successfully delivered;
  8. voters who attempted access, submitted a ballot, opened a support case or took no recorded action;
  9. challenges by decision type, outcomes and median resolution time;
  10. changes after provisional publication and after final freeze;
  11. provisional ballots or quarantined records and their final disposition; and
  12. votes accepted, duplicates prevented, ballots excluded and certification exceptions.

The report should explain each reason category and identify the rule version. It should state what it cannot determine. If a person completed biometrics but never opened the ballot, the institution may classify the case as no recorded ballot access; it should not speculate about motive. If a support ticket says a credential failed, that can be classified as reported access failure, with the final remedy separately stated.

An independent assurer should test samples and totals against protected evidence. The assurer should be able to say whether every person in the 581-to-548-to-484 funnel has one final status and whether the broader eligible-member denominator reconciles to the voter register.

The report's purpose is not to relitigate every administrative judgment in public. It is to prove that every judgment existed, had a reason and could be challenged in time.

What the 2025 record supports, and what it does not

The public evidence supports several firm conclusions. AFRINIC's June design used documentary authority and identity requirements for in-person voting. Questions arose around powers of attorney. The receiver annulled the election while stating that investigations had not reached final conclusions. The September replacement used direct designated voters, identity documentation, biometric registration and online voting. The published statistics reported 581 voters, 548 completed biometric registrations and 484 votes.

The same evidence supports a more limited institutional conclusion: the published record does not supply the complete excluded-voter denominator, reasons, appeals and final handling needed to explain every gap. It shows stages but not all transitions. The 33 and 64 differences are facts; their composition is not.

The record does not support accusing any named person or organisation of voter fraud. It does not establish that every power of attorney was defective. It does not prove that the September register was inaccurate, that biometric verification was biased, or that the 64 verified non-voters were blocked. It does not prove that a different exclusion count would have changed any seat.

These limits do not weaken the analysis. They define it. The governance failure is not that one preferred allegation cannot be proved from public material. It is that a high-stakes restoration election should produce enough structured evidence to separate allegation, administrative decision, technical failure and voter choice.

The right denominator is a continuity control

AFRINIC's board election concerned more than corporate ceremony. The board would supervise an institution whose records, registration services and routing-adjacent functions matter to networks across a large region. A legitimacy dispute could prolong exceptional control, litigation and uncertainty. Election evidence was therefore part of service continuity.

A board elected through a well-reconciled franchise can make decisions with greater confidence. Members who lose can inspect the rules and know whether their organisation had a fair route to participate. Courts receive narrower disputes with contemporaneous records. External coordinating bodies can assess continuity without substituting their own preferences for member choice.

The opposite is also true. If excluded voters are invisible, every later dispute can expand. A single complaint becomes a theory about the whole roll. Administrators defend the result with accepted-vote totals that do not answer access questions. Critics fill gaps with allegation. The institution spends more time proving legitimacy after the fact than it would have spent preserving evidence before the vote.

Counting exclusion is not an admission that exclusion was wrongful. It is the basic control that makes lawful exclusion distinguishable from disenfranchisement.

Sources and analytical limits

AFRINIC's 2020 constitution supplies the company and membership framework, voting rights and proxy provisions. It establishes formal categories but does not prove how any individual member or voter was classified in 2025.

The June access requirements are drawn from AFRINIC's 18 June 2025 communiqué, the 2025 board-election page and related published voting guidance. These materials state announced requirements; they are not an independent observation log from the polling station.

The treatment of the annulled June election relies on the receiver's 15 July 2025 notice. The notice is evidence of the receiver's stated concerns, referral and remedy. Its express acknowledgement that investigations lacked final conclusions is preserved; it is not used as proof of fraud or of the reach of any irregularity.

The replacement design is documented in the 2025 election guidelines, the designation process, the voters register, the privacy notice and the published election statistics. They establish announced procedures and reported aggregate numbers, not complete independent proof of execution.

ICANN's 25 June 2025 letter is used only as evidence of questions raised about the June arrangements. It does not decide the validity of a ballot, document or member. No complete member denominator, rejected-voter list, polling-station log, identity-provider exception file, platform incident record, appeal register, police file or sealed court evidence was available for this analysis.

The proposed exclusion record, provisional-ballot route and assurance controls are governance recommendations. Their implementation would need to comply with Mauritius law, AFRINIC's constitution, applicable court directions, data-protection requirements and ballot-secrecy safeguards. The article does not determine any individual's voting right or the legal validity of the September result.