Summary

  • Between late February and final recognition on 8 April 2005, AFRINIC made primary numbering-resource allocation evaluations while “existing” or “other” RIRs supplied a temporary second opinion or overview. The public record identifies the reviewer class and the recognition-based sunset, but not a case reviewer, a quantitative review trigger, or a request-level audit trail.
  • A June 2004 design document spoke of “second approval by RIRs,” whereas the operating record used “second opinion” and “overview process.” Because no published disagreement procedure explains vetoes, escalation, notice, cure, appeal, override or stays, the decision path cannot be reconstructed when the first and second views diverged.
  • The period was 46 calendar days if counted from the NRO’s 21 February start record, or 45 if counted from AFRINIC’s announced 22 February effective date. Those figures are calendar bounds, not official performance measures. Recognition on 8 April met the stated sunset condition, while the treatment of open requests through an anticipated 11 April activation remains unknown.
  • The control demonstrated bounded operating exposure: AFRINIC could run the primary desk and production systems while accepting a short external check as a broader readiness review continued. It did not establish a perfect concurrence rate, prove that every allocation was correct, or grant sovereign, legislative, police, punitive, prosecutorial, confiscatory or transnational authority to either the checked registry or the checking registries.

The request at the centre of the control

In late February 2005, the critical administrative object was not a ceremonial transfer. It was a numbering-resource allocation decision moving through AFRINIC’s own desk. The public record supports a spare but consequential sequence. A request reached AFRINIC; AFRINIC staff and registration systems occupied the primary role; AFRINIC evaluated the request; and an experienced peer layer drawn from other regional Internet registries supplied a second opinion. The record does not name an applicant, a prefix, an approval, a refusal or a disputed outcome, so none should be imagined.

The concrete fact is the path itself: primary judgment had moved to AFRINIC while a temporary check remained.

That structure matters because it separates two risks that are often confused during an institutional handover. One is the risk that a new operating desk never gets to exercise judgment because its predecessors continue to make the real decisions. The other is the risk that the new desk exercises judgment without any short-term mechanism for detecting an evidential or policy error. A second opinion can answer both problems. It leaves the first evaluation with the incoming operator, making its capacity observable, while retaining another set of eyes during the most exposed part of the transition.

The arrangement also created a less comfortable risk: a nominally independent desk could remain subject to an opaque external veto. Whether that occurred cannot be established. The language of the documents does not settle the question, and no case file shows what happened if the two views diverged. The control therefore deserves neither automatic celebration nor an invented scandal. It deserves an audit-minded reading that distinguishes what its outline accomplished from what its missing mechanics prevent the public from knowing.

Three stages, three different allocations of responsibility

The transition’s sequence can be located without turning this report into a general history of AFRINIC’s operational handover. Before the interval examined here, AFRINIC had participated in co-evaluation of African resource requests beginning in September 2004. Co-evaluation meant a joint first-stage process; it is not the same control as a primary AFRINIC evaluation followed by a second opinion. The distinction is essential. Joint work can train staff and expose them to live cases, but it makes it difficult to observe which institution actually owns the first judgment.

The February 2005 arrangement changed that relationship. AFRINIC’s staff and systems took the primary role and AFRINIC described itself as operating with a second opinion from other RIRs. The updated application placed the second-opinion process alongside transferred operational functions, including resource-record administration, reverse-DNS maintenance, WHOIS and customer interfaces. Those functions are context, not the subject. Their relevance is that the second opinion did not substitute for AFRINIC’s operating desk. Primary operation and peer checking coexisted.

The next stage was described as operation without the second opinion. On 23 February, the chief executive said that stage was scheduled to arrive in less than two months. This forward-looking statement corroborates the temporary character of the control, but it does not identify the last reviewed request. The useful three-part distinction is therefore compact: earlier joint co-evaluation; then AFRINIC’s primary evaluation under a second opinion; then AFRINIC’s operation without that transitional opinion. Blurring those stages would either understate the incoming desk’s responsibility or exaggerate the reviewer’s role.

A transition control, not a transfer of rule

The second opinion was legitimate to the extent that it served thin coordination. A numbering registry keeps a shared ledger useful by protecting uniqueness, applying published administrative criteria, maintaining accurate records and preserving operational continuity. An experienced peer can check whether evidence was handled consistently, whether a policy was applied coherently and whether a proposed record creates an avoidable ledger risk. Those are administrative and technical functions. They are valuable precisely because errors in the common record can affect operators beyond the desk that made them.

No part of that work creates government. AFRINIC and the reviewing RIRs were private bookkeepers and technical coordinators. They held no sovereign, legislative, regulatory, police, punitive, prosecutorial, confiscatory or transnational public-law authority. The checking registry was not an imperial superior, and the checked registry was not a new ruler awaiting coronation. Institutional recognition could confirm that coordination criteria had been met; it could not manufacture jurisdiction, ownership of Internet reality or a power to punish.

This boundary changes how the review should be judged. The proper questions are whether an administrative evaluation was traceable, independently checked, reasoned, correctable if error was detected and terminated under a known condition. The wrong questions are which institution enjoyed the higher political rank or whether a peer’s concurrence consecrated the decision. Peer review can improve a ledger decision. It cannot turn private agreement into public law.

The same boundary protects applicants. A registry may ask for evidence needed to apply an allocation policy and may decline to record a request that does not satisfy that policy. It may not disguise punishment, prosecution or confiscation as technical review. A second opinion should narrow error and expose reasons; it should never become an unreviewable device for imposing sanctions through delay, secrecy or withdrawal of an operational record.

The trigger that the record does not disclose

A control cannot be fully evaluated until one knows what activates it. Here, the surviving wording associates the second opinion with AFRINIC’s allocation decisions. That is enough to identify the control surface but not enough to establish its trigger. No published record specifies a minimum prefix size, financial value, request category, risk score, materiality test, exception class, conflict indicator, random sample or escalation threshold. Nor does the broad wording safely prove that every request received a second review.

The absence is easy to obscure because a different threshold appears in the transition material. Policy harmonisation work supported a /22 minimum IPv4 allocation in the ARIN and RIPE NCC service areas in Africa. That /22 was a substantive allocation-policy threshold: it concerned the size of an allocation under the relevant policy. It was not a published trigger for second-opinion review. Treating the /22 as the review threshold would convert one documented rule into another rule for which there is no evidence.

The difference is not technical trivia. If every allocation decision was reviewed, the control offered consistent coverage but also duplicated work across the full request stream. If only larger, novel, conflicted or weakly evidenced requests were reviewed, the control may have concentrated effort efficiently, but someone had to classify cases and could exercise hidden discretion at that gate. If cases were sampled, the sampling method would affect what error signal the transition produced. None of those designs can be selected from the public record.

This missing trigger also limits any claim about readiness. Exposure to a peer check is meaningful, but the depth of that exposure depends on coverage. A desk whose entire allocation stream was reviewed faced a different test from one whose cases were selectively referred. Without the activation rule or case count, the period demonstrates that a control existed around primary evaluation, not how much of the decision population it tested.

A reviewer class without a case reviewer

The institutional reviewer is described only in plural and generic terms: “existing RIRs” or “other RIRs.” IANA’s recognition report names APNIC, ARIN, LACNIC and RIPE NCC as the existing regional registries supporting AFRINIC through the NRO. The historical service transition in Africa chiefly involved APNIC, ARIN and RIPE NCC. Those facts identify the surrounding institutional field. They do not map a particular allocation request to a particular registry, team or hostmaster.

No public reviewer roster, rotation, case-assignment rule or conflict-of-interest procedure survives in the record considered here. There is no basis for saying that ARIN reviewed one category, RIPE NCC another, APNIC a third, or LACNIC a specific case. There is likewise no basis for naming an individual reviewer or reconstructing a bilateral workflow. The most precise formulation remains the least dramatic one: an established-RIR reviewer class participated, while case-level identity is unknown.

That gap weakens traceability. An auditable second opinion normally needs to show who was assigned, when the file was received, which materials were examined, whether the reviewer had a conflict, what conclusion was reached and how that conclusion affected the first evaluation. Confidential applicant evidence need not be published. But the existence of a reviewer assignment and its control metadata can be recorded without exposing the applicant’s network plan, customer information or operational secrets.

Reviewer identity matters for institutional learning as well as accountability. Different registries may have brought different policy traditions, service-area experience and internal practices to the check. Without assignment data, one cannot tell whether review was balanced across peers, concentrated in a predecessor registry, delayed by workload or shaped by inconsistent interpretations. The record therefore supports experience at the class level but not consistency at the case level.

“Second approval,” “second opinion” and “overview” are not synonyms

The sharpest clue to unresolved decision rights lies in the wording. A June 2004 transition presentation described Phase 2 as uniform policy and evaluation with “second approval by RIRs.” In February and March 2005, the operating descriptions instead referred to a “second opinion,” including an “overview process.” A later planned stage involved independent evaluation with audit by RIRs. These terms describe different possible positions in a decision chain.

“Second approval” can imply that two affirmative decisions were required before the allocation could proceed. Under that reading, the reviewer could exercise a formal veto, at least during the transition. “Second opinion” can imply advice delivered before a final decision, influential but not necessarily binding. “Overview” may suggest supervisory visibility across decisions rather than a second signature on each one. “Audit” commonly suggests review after a decision or across a sample. The documents do not reconcile these possibilities.

It would be equally unsafe to assume that the stronger design phrase controlled every live case or that the softer operating phrase made the reviewer purely advisory. Plans can change when implemented; public announcements can simplify; institutions can use different words for the same internal rule. Only a workflow document or case trail could show whether the peer’s view was a precondition, a recommendation, an after-the-fact check or some combination. No such material establishes the answer.

The uncertainty goes to the centre of accountability. If the reviewer could block a proposed allocation, responsibility for a refusal or delay was shared even though AFRINIC occupied the primary desk. If AFRINIC could override the peer, then the second opinion created an error signal but left final responsibility at AFRINIC. If the review occurred only after entry, it could inform correction and learning but might not prevent an immediate mistake. A transition control should state these rights in verbs, not leave them to institutional titles.

The disagreement path cannot be reconstructed

No disagreement is documented at request level, and none should be invented. Yet a second opinion is most valuable analytically when it differs from the first. Agreement may confirm consistent policy application; disagreement reveals whether the control can locate the disputed issue, improve reasoning and correct error without hiding power. The public record offers no tie-breaker, escalation body, response deadline, reason-giving standard, notice to the applicant, opportunity to cure evidence, appeal, override, stay rule or method for recording dissent.

This absence has two consequences. First, empirical effectiveness cannot be measured. There is no count of agreements, disagreements, changed outcomes, false approvals, false refusals, processing delays, complaints, reviewer workload or error rates. A claim that the control caught mistakes, produced perfect concurrence or caused no delay would exceed the evidence. So would a claim that it failed. The record exposes the existence and duration of a safety layer, not its measured performance.

Second, the allocation of decision power remains unreconstructable. Suppose the first evaluator and peer applied the same policy differently. Did the request pause? Did one registry’s view prevail? Was another expert asked? Did the applicant learn why additional evidence was needed? Could AFRINIC proceed over objection, and if so, was the dissent retained? These are not allegations about what happened. They are the minimal questions by which any two-reviewer control is understood.

Opacity here can produce hidden veto as easily as superficial review. If the second view silently controlled, applicants might experience delay without knowing which decision or reason they had to address. If the second view carried no defined consequence, the reviewer might provide ceremonial comfort without changing a weak decision. If disagreements were resolved informally, experienced staff may have reached sensible outcomes, but the institution would retain little durable evidence of how judgment improved. Each possibility is plausible in the abstract; none is established for the 2005 cases.

A condition-bounded interval with two start dates

The public record does identify the control’s termination condition more clearly than its activation rule. The NRO said the second-opinion overview would continue until AFRINIC received final recognition. That is a condition-based sunset, not an indefinite promise of guardianship. The chief executive’s statement that the next phase would run without the second opinion and was scheduled for less than two months supports the same temporary design.

The starting point must be presented with a one-day discrepancy. The NRO progress letter said AFRINIC staff and registration systems would take the primary role beginning 21 February 2005. AFRINIC’s public announcement said Phase II took effect on 22 February. Both dates belong in an accurate account. Collapsing them into a single apparently certain timestamp would hide a difference in the contemporaneous sources.

ICANN granted final recognition on 8 April through resolutions 05.25 and 05.26, recording that AFRINIC had reached the conclusion of its transition plan and met the requirements for recognition. Because final recognition was the published sunset condition, 8 April is the documented date on which that condition was satisfied. It follows that the interval has two defensible calendar bounds: 46 days measured from 21 February to 8 April, or 45 measured from 22 February to 8 April.

Those numbers are calculations, not institutional performance metrics. They do not reveal how many cases entered the control, how long any review took or when the last peer opinion was delivered. The record also does not identify the exact hour of termination. It cannot show whether a request already under review on 8 April completed under the old arrangement or shifted immediately to AFRINIC-only treatment.

An anticipated 11 April date adds a further boundary without resolving it. The NRO had expected full operational activation on 11 April after anticipated ICANN action during 4–8 April. Recognition occurred on 8 April, but no published rule establishes whether 11 April became the practical cutover for open requests. The responsible conclusion is narrow: 8 April met the stated condition, while the treatment of in-flight work between recognition and the anticipated activation remains unknown.

What readiness the interval did demonstrate

The strongest readiness inference is institutional rather than statistical. AFRINIC could receive requests directly, operate production registration and WHOIS systems, carry out the primary allocation evaluation and expose its judgment to a temporary external check. A new desk that can perform real work under observation provides more evidence of capacity than one that participates indefinitely in joint evaluation without owning the first decision. The control made operating responsibility visible while keeping a short error-detection layer around it.

The broader recognition assessment supplied separate readiness evidence. The NRO reported that internal registration, external WHOIS and accounting systems were in production and that facilities in Pretoria and Mauritius had been reviewed. IANA described staff training, engineering exchanges and collaboration on legal, financial, administrative and registration matters, and concluded that the transition plan was satisfactorily executed and virtually complete.

ICP-2’s criteria included regional coverage and support, documented policy procedures, neutrality, technical expertise, adherence to global policies, an activity plan, funding, record keeping and confidentiality.

These inputs must not be collapsed into a claim that the second opinion alone caused recognition. Systems, facilities, staffing, institutional planning and documentary criteria formed a wider bundle. Nor did recognition supply the missing allocation-review metrics. ICP-2 did not state a numerical concurrence threshold, and no record gives a second-opinion pass rate. The peer check was one part of supervised operating exposure, not a statistical proof of perfect decisions.

Readiness, on this evidence, meant the capacity to run the desk, accept a bounded check and complete a broader interoperation assessment. It did not mean infallibility. No institution becomes error-free because its peers express confidence, and no private registry acquires public-law authority because another private body recognises its operational status. The useful evidence is the disciplined handover of administrative work, bounded by the uncertainty of an unpublished case ledger.

The strongest case for a lean informal check

There is a serious defence of the sparse public design. The transition lasted only weeks. Incumbent registries had trained AFRINIC staff and supported the handover. Experienced hostmasters working across established professional relationships may have been able to resolve interpretive differences quickly and informally. A thick public procedure could have consumed attention precisely when the new desk needed to keep service moving. Publishing request files would also have threatened legitimate confidentiality around applicant evidence and network planning.

On that view, the second opinion was a lightweight safety device, not a permanent tribunal. Its success may have lain in quiet expert contact: ask a peer to test the conclusion, correct an obvious inconsistency if one appears, and remove the extra layer as soon as recognition is complete. A short transition does not necessarily need a bureaucracy built to last for years. It may need trust, speed and a clear exit.

That case has force. The answer is not to demand public disclosure of confidential requests or to confuse administrative review with litigation. It is to separate case contents from control metadata. The institutions could have disclosed the trigger, assignment method, reviewer role, response target, decision rights, reason codes, disagreement count and in-flight sunset rule without naming an applicant or revealing sensitive evidence. Those fields would have made the control inspectable while preserving its lean character.

Informality is most defensible when the surrounding accountability is precise. A short-lived arrangement can justify fewer layers, but its short life also makes later reconstruction harder: staff disperse, recollections fade and private messages may not form a durable record. The cost of a small audit trail is low relative to the uncertainty created when even the reviewer’s case identity and disagreement effect cannot be recovered publicly.

Operational stakes without back-projected history

The immediate stakes in 2005 were practical. An inconsistent allocation decision could produce an inaccurate registry entry, impair policy consistency or complicate applicant planning. An unjustified refusal could delay network deployment; a weak approval could create avoidable record and coordination risk. A second review could reduce single-evaluator error, while a poorly specified one could duplicate documentation, slow processing or place hidden discretion behind an unnamed peer.

Later IPv4 scarcity makes allocation records economically more valuable and makes auditability easier to appreciate. It does not prove that later scarcity, leasing disputes or later allocation failures motivated the 2005 participants. Those developments must not be projected backward as a secret purpose, and later control research is not evidence that this particular second-opinion period failed. The proper use of later experience is conceptual: it shows why the chain from request to evaluation, approval, inventory movement and public record should be independently verifiable.

LARUS treats registry decisions as infrastructure risks because an administrative choice can quietly affect operations even though the registry has no sovereign foundation. NRS makes the structural boundary equally clear: regional registry cooperation is coordination without sovereign enforcement, and expansion beyond that contractual function makes governance fragile. BTW’s allocation-control work treats the full decision chain as a control surface that should leave evidence. These are first-class lessons about consequence and design. They do not supply a missing 2005 case or performance statistic.

The economic mechanism is therefore bounded. Independent checking can catch inconsistent evidence or policy application before a ledger entry becomes operationally consequential. Missing reviewer assignment, disagreement rules and metrics can instead produce unmeasurable discretion and delay. Both effects are possible in a two-reviewer design; the surviving record proves the existence of the design outline, not which effect dominated in practice.

A ledger can be checked without becoming a throne

The controlling doctrine is straightforward. A registry may record, coordinate, protect uniqueness and keep a thin common layer accurate and auditable. It may not rule. Institutional deference is not authorship of the operational reality described by the ledger. Recognition is not sovereignty. Cooperation among RIRs is not jurisdiction over persons, property or states. Administrative review is not policing, prosecution, punishment or confiscation.

Applied to the second-opinion interval, this doctrine supports the check while denying its inflation. AFRINIC’s primary evaluator could test evidence and apply allocation policy. A peer could identify inconsistency and ask for a reasoned reconsideration. If error were found, the record could be corrected through an administrative process. None of those acts required a claim of public supremacy. Their legitimacy depended on accuracy, traceability, contractual coordination and the continuing ability to audit the common record.

The doctrine also explains why termination matters. A transitional reviewer that persists after its condition has been met can turn assistance into unaccountable dependency. A clean sunset respects the incoming operator’s responsibility while preserving the evidence created during the supervised interval. The durable model is not permanent guardianship by older registries. It is a temporary, inspectable check that ends when its stated condition is satisfied.

The public record shows that the control was designed to end. It does not show enough to audit how it operated case by case. That is the balanced finding supported by the evidence: sensible architecture at the level of roles and duration; incomplete accountability at the level of trigger, assignment, disagreement and measurement. The ledger could be checked. Neither checker nor checked could rule.