Summary

  • Beginning on 21 February 2005, AFRINIC staff and production systems took the primary role in handling regional registry services for African Local Internet Registries and end users. But this was a supervised operational handover, not a moment when every legal, custodial, accreditation and resource-allocation power moved at once.
  • Final ICANN recognition followed 46 elapsed days later, on 8 April. A first post-recognition IPv4 and ASN package was announced on 14 April, AFRINIC joined the Number Resource Organization by an instrument dated 25 April, and the incumbent registries signed later authority-transfer letters on 26–27 April. These dates describe different institutional acts.
  • The surviving records leave material gaps. They do not disclose how the “second opinion” procedure worked in individual cases, how often peers intervened, which pools supplied every Phase II allocation, or whether every APNIC, ARIN and RIPE NCC contract and record migrated on identical terms. A one-day disagreement between 21 and 22 February also remains unresolved.
  • The sequence supports the strongest case for staged accreditation: AFRINIC demonstrated production capability under supervision before receiving independent standing in the recognised RIR system. It also fixes the limit of that case. Recognition did not switch on systems that were already running, confer ownership of member networks, control routing, or grant a private registry sovereign punitive power.

At some point on the February cutover day, an African network seeking registry service stopped being merely a future customer of AFRINIC. The request path had changed. AFRINIC’s hostmasters were taking requests; its registration, WHOIS and accounting systems were in production; and, for the portion of the region previously served by RIPE NCC, the old customer portal was no longer the place to work. This was not a conference resolution anticipating a future service. It was an operating change experienced through interfaces, records and staff.

Yet the institution at the centre of that change was not finally recognised by ICANN as an independent regional internet registry. The incumbent RIRs continued to provide a second opinion on allocation decisions. At least for African Local Internet Registries inherited from RIPE NCC, legal membership and the standard service agreement remained with RIPE NCC until final recognition. Some database records moved, some were copied, and some routing-registry objects stayed where they were. Later instruments would address recognition, number-resource inventory, NRO membership and formal authority transfer.

That is the central fact of 21 February 2005: AFRINIC was running the service before all of the institutions around the service had finished changing their own records and relationships. The sequence is easy to flatten into a ceremonial origin story in which one date creates an organisation, an authority and a functioning registry. The evidence instead presents a distributed cutover. Different parts of the arrangement changed at different times because they were different things.

A working service before the final label

The contemporaneous joint progress report from the NRO chair and AFRINIC’s chief executive says that AFRINIC’s systems were in production and serving registrants. Beginning on 21 February, according to that report, AFRINIC staff and systems took the primary role in providing registry service. Existing RIRs would maintain a second-opinion overview until final recognition. ICANN’s public-comment notice of 14 March later described APNIC, ARIN and RIPE NCC—the three incumbent registries that had served different parts of Africa—as having transferred all RIR services on 21 February.

RIPE NCC’s own transition record also places completion of its operational transitions on that date.

Two other institutional records use 22 February. An AFRINIC mailing-list announcement and an APNIC archive call that the effective beginning of Phase II, described as full RIR operation with second opinion. The one-day difference is not evidence for two separate handovers. Both date clusters describe the same transition stage. But neither the available records nor a cutover log explains whether the difference arose from time zones, the sequence of technical changes, publication lag or inconsistent wording.

The defensible institutional date for this event is 21 February because it appears in the joint notice, ICANN’s later summary and RIPE NCC’s detailed account. The 22 February wording must remain beside it as an unresolved documentary discrepancy.

That care over one day is not pedantry. Dates become dangerous when they are made to carry powers they did not record. If 21 February is called the date AFRINIC “became the RIR” without qualification, a reader can easily infer that corporate personality, production service, resource custody, peer independence, ICANN recognition and NRO membership arrived in one package. If the date is discarded in favour of the later recognition ceremony, the opposite distortion follows: real service work disappears until a board records its approval. The evidence supports neither simplification.

The service itself was tangible. New requests went to AFRINIC hostmasters for evaluation. AFRINIC’s internal registration system, external WHOIS publication and accounting functions were live. For RIPE-served African LIRs, requests were redirected and the RIPE NCC portal ceased to be available. Registry data and reverse-DNS administration went through planned transition procedures. These are operational facts: where a request was sent, which staff handled it, which database published the record and which organisation faced the customer.

They are also bounded facts. “Primary role” does not mean exclusive control over every decision. “All RIR services” does not establish that every contract, resource pool, object, DNS delegation and custodial instrument changed at exactly the same instant. “Fully functional,” as used by ICANN while recognition was still pending, cannot mean that every institutional dependency had ended. The source record itself supplies the boundaries because the same institutions that described full operation also described continuing second opinion and later formal acts.

The cutover at database level

The RIPE NCC account provides the clearest surviving view of what a service handover meant on the ground. It describes request redirection, the loss of access to the former portal for African LIRs, changes to WHOIS handling and a freeze around reverse-zone migration. It also records exceptions. Certain objects were copied rather than simply removed, while routing-registry material remained in the RIPE Database because AFRINIC did not then operate a routing registry.

The transition page identifies two ranges of aut-num objects that were copied: 30980 through 30999, and 34515 through 34519. The first inclusive range contains 20 objects and the second five, for a total of 25. That number is not an allocation count and not a measure of the whole transfer. The objects remained available in RIPE’s database for routing-registry use. Other aut-num objects and provider-independent assignments remained registered there under the arrangement described. The detail matters because it shows that even one broad word—“database”—contained several functions and migration rules.

A registry database can publish recognised resource information. A routing registry can publish intended routing policy. Reverse DNS delegates administrative control over address-to-name mapping. An account system establishes who is billed and through which relationship. A request portal determines where operators submit work. Moving one does not automatically move all the others, and none of them physically makes a router accept a route. The February transition coordinated these surfaces without pretending they were interchangeable.

The granular RIPE record cannot be projected wholesale onto the African LIRs previously served by APNIC or ARIN. ICANN’s aggregate notice says the three incumbent service RIRs transferred all RIR services. APNIC’s archive says African LIRs and their allocated resources transferred. But comparable public inventories for APNIC and ARIN—covering portal rules, contracts, object types, DNS steps, individual cases and effective timestamps—were not located in the sealed record. The gap does not erase the high-confidence finding that the service handover occurred.

It prevents a stronger assertion that every predecessor carried it out through identical legal and technical mechanics.

That distinction is especially important when an institutional summary uses total language. The phrase “all RIR services” is strong evidence of what ICANN recorded as the operational result. It is not a substitute for the missing schedules. The later NRO account of authority-transfer letters, including registration and accounting documentation and DNS administration, confirms that some formal custodial work remained to be completed in April. A broad summary and a detailed implementation record can both be true if they describe different layers.

Six layers, not one authority

The handover becomes intelligible when the word “authority” is divided into the powers people actually exercised.

First was running service. AFRINIC staff received and evaluated resource requests, maintained the customer interface, processed registry and accounting records, and published service data. This was the layer that clearly changed on 21 February. Its proof lies in live systems and redirected work, not in ceremonial language.

Second was peer supervision. The existing RIRs retained a second-opinion overview of AFRINIC allocation decisions until final recognition. They also supplied technical support, training and transition monitoring. This did not make them the primary service desk after the handover, but it prevents the claim that AFRINIC already exercised wholly unsupervised allocation judgment. The public corpus does not reveal whether second opinion meant mandatory concurrence, a binding veto, consultation, post-decision assurance or some combination. It does not name reviewers, give response times or show an escalation procedure.

The existence of supervision is high-confidence; its practical intensity remains unknown.

Third was number-resource authority and inventory. IANA administered top-level pools for RIRs. Incumbent registries retained transitional roles around pools, legal relationships or custody, while AFRINIC processed and recorded allocations under supervision. ICANN announced on 14 April that AFRINIC had received its first post-recognition allocation including IP addresses and autonomous system numbers. The present IANA IPv4 registry records 41/8 for AFRINIC in April 2005, and the ASN registry records 36864 through 37887 in April 2005. Those are later, specific inventory milestones.

There is a complication that should stop any neat claim that AFRINIC had no IANA-associated resource before recognition. The current IANA IPv6 unicast registry lists 2001:4200::/23 for AFRINIC with a date of 1 June 2004. The sealed evidence does not explain how that line relates to the contemporaneous description of the April 2005 package as the first post-recognition allocation. The safe reading is narrow: 14 April marks the announced post-recognition package and is supported by April IPv4 and ASN records; it is not proof that every class of IANA-linked inventory first appeared then.

Fourth was accreditation within the recognised RIR system. ICANN’s board granted provisional approval on 30 September 2004. That decision found a reasonable basis for eventual recognition, required completion of the transition and a supplemented application, and directed continued work among AFRINIC, the incumbent RIRs and the NRO. After an IANA evaluation against ten ICP-2 criteria, the board gave final approval on 8 April 2005. This changed AFRINIC’s standing in the globally coordinated RIR framework. It did not incorporate the company, write its software, or cause packets to flow.

Fifth was domestic legal personality. AFRINIC’s updated application says the organisation had been incorporated in Mauritius as a not-for-profit in February 2004. That is an applicant statement and no underlying certificate was located in the sealed research. A first-class NRS history page instead uses wording that AFRINIC was incorporated in April 2005. The two published accounts conflict. Without the Mauritius registry instrument, the exact corporate chronology must remain attributed: the applicant put incorporation in February 2004; NRS used April 2005 wording.

Either way, corporate existence is conceptually distinct from ICANN recognition. One arises through applicable domestic law; the other admits an operator to an institutional coordination role.

Sixth was routing and network operation. Independent network operators originated and accepted routes, operated their equipment and carried continuity risk. A registry record can be relied upon, ignored, disputed or used as an input by those operators. It can make coordination far easier and conflicting claims far harder. It still does not physically command BGP, own member routers or hold title to the networks registered in its database. No reviewed instrument transferred African territory, criminal jurisdiction, police power or general regulatory sovereignty to AFRINIC.

These layers interacted, but they did not fuse. AFRINIC’s production service gave operators a working regional interface. Peer supervision reduced the risk of an inexperienced registry making unchecked decisions. IANA-facing inventory enabled recognised allocation from global pools. ICANN accreditation established status in the RIR system. Mauritius law supplied the corporate container for employment, contracts, membership and governance. Network operators retained practical routing control. The NRO provided peer coordination, not a continental government.

Calling AFRINIC a ledger and coordinator is therefore neither insult nor euphemism. The ledger is consequential. Errors in recognised records can delay transactions, complicate due diligence, misdirect operational reliance or raise the cost of proving a holder’s status. A registry can exercise contractual rights and control service access within the bounds of its agreements. But consequential coordination is not sovereign punishment.

AFRINIC was never made the police, prosecutor or judge of African network operators by the February handover, and it did not become the owner of their addresses in the sense of owning their networks or economic activity.

The 46-day interval

From the selected operational date of 21 February to final recognition on 8 April, 46 elapsed calendar days passed. Using the alternative 22 February wording would produce 45 days. That interval was not an empty waiting room. AFRINIC was processing work while the institutional system observed a supervised operator.

The sequence had begun well before February. AFRINIC’s application dates growing momentum for an African registry from 1997. It says the Mauritius entity was incorporated in February 2004, although the underlying certificate is missing and NRS’s later chronology conflicts. In May 2004, AFRINIC held a public policy meeting and annual general meeting in Dakar; the application says a constitution was adopted and initial resource policies were processed. In September 2004, the recognition application entered the formal process and AFRINIC began co-evaluating African resource requests with incumbent RIRs.

Provisional approval followed on 30 September.

By January 2005, NRO Executive Council representatives had visited AFRINIC’s facilities in Mauritius, according to the application. RIPE NCC says its African LIRs began receiving AFRINIC contracts by courier. Yet those LIRs remained legally RIPE NCC members pending final recognition, and their assignments and registrations remained under the RIPE NCC Standard Service Agreement. For LIRs that had already paid RIPE NCC for 2005, the transition was designed to avoid double payment. Contract migration and operational migration were deliberately offset rather than forced into a single instant.

On 21 February the operational centre moved. On 5 March, AFRINIC’s updated application represented that all operational aspects were under its management while allocation decisions remained subject to second opinion. On 14 March, ICANN opened a 21-day public-comment period and described AFRINIC as fully functional. The application was still pending. This coexistence—fully functioning service and unfinished accreditation—is not an anomaly once function and recognition are understood as separate objects.

On 8 April the ICANN board finally approved and recognised AFRINIC to provide IP registration and related services for the African region. Six days later, on 14 April, ICANN announced the post-recognition allocation package. From 21 February that was 52 elapsed days. The April IPv4 and ASN entries locate an important direct-inventory change but, because of the 2004 IPv6 entry, cannot safely be called the first IANA association of every kind.

The sequence continued. AFRINIC’s joinder to the NRO memorandum is dated 25 April, 63 elapsed days after 21 February. That contractual instrument gave AFRINIC rights and duties as an NRO member. It did not retroactively cause the February systems to operate. During the AFRINIC-2 meeting on 26–27 April, according to the NRO, APNIC, ARIN and RIPE NCC signed letters transferring formal authority for their former African service areas. The NRO’s 27 April account names registration and accounting documentation and DNS administration, saying the letters completed custodial and registry processes.

The report date was 65 elapsed days after the service start; the precise effective moment of each letter is unknown because the letter texts were not located.

Chronology here is a form of institutional accounting. A body can be legally incorporated before it is accredited, operate before it is fully independent, receive inventory on one timetable, join a coordination organisation on another and complete custody through later instruments. Each act answers a different question. Which staff process a request? Who reviews the judgment? Which pool supplies the numbers? Which body recognises the registry? Which corporation signs the contract? Who publishes the data? Which network accepts the route? The historical record becomes misleading only when one answer is made to stand for all of them.

Why operators and holders cared

For an operator, the value of the February handover was not abstract regional symbolism. It was a service pathway. Requests needed somewhere to go. WHOIS data needed to remain discoverable. Reverse DNS needed an administrator. Accounts, contracts and records had to be changed without forcing networks to renumber or leaving holders unable to show recognised status.

The staged approach could reduce discontinuity. AFRINIC could acquire operational responsibility while experienced peers still checked allocation decisions. RIPE NCC could direct new work to AFRINIC while maintaining a temporary legal relationship with its African members. Existing payments could be respected so that the institutional move did not become double billing. Routing-registry objects could remain at RIPE while the new registry lacked that particular service, rather than being removed merely to make the organisational diagram look clean.

Those same safeguards created split accountability. An LIR could be served by AFRINIC while its contract remained with RIPE NCC. A request could be evaluated by AFRINIC while another RIR supplied second opinion. A record could appear in one registry system while a routing-policy object remained in another. If a decision were delayed or challenged, the public documents do not show who had the final operational obligation, what escalation applied or how an operator would distinguish assurance from veto.

No case ledger was located. There is no public denominator for Phase II requests, approvals, rejections, modifications, delays or reversals. There are no published counts of peer concurrence or intervention, no named reviewers and no processing-time series. It would therefore be equally speculative to say the second-opinion mechanism dominated AFRINIC’s decisions or to say it was ceremonial and never used. The record proves the arrangement, not its frequency or effect.

Nor does the absence of outage data prove a flawless transfer. RIPE NCC described freezes and referrals intended to minimise operational impact, while the institutional sources described a functioning service. But no availability figures, queue-depth records, DNS error rates, incident reports or comprehensive operator testimony appear in the sealed evidence. The correct claim is that the transition was planned and the service was reported as operating—not that every user experienced zero disruption.

The economic mechanism is straightforward. Registry reliability lowers the cost of establishing a recognised claim, obtaining additional resources, maintaining reverse DNS and keeping contact records coherent. Confused authority raises transaction costs: holders must determine which organisation can accept a request, which one reviews it, which contract applies and which database counterparties will trust. A staged handover can manage those costs if its interfaces and responsibilities are clear. The same staging can increase them if institutions speak as though “service,” “authority” and “recognition” are synonyms.

The strongest case for recognition

The operational-first sequence is not an argument that accreditation was useless. The strongest contrary reading is that it shows a responsible accreditation process working as designed.

On that account, provisional approval created a controlled path rather than an irreversible endorsement. Incumbent registries trained AFRINIC and co-evaluated requests. NRO representatives inspected facilities. Phase II placed staff, databases and customer interfaces into production while preserving second opinion. IANA then examined ten ICP-2 criteria, including regional scope, LIR support, policy development, neutrality, technical capability, global-policy adherence, planning, funding, recordkeeping and confidentiality. Final recognition followed observed operation rather than preceding it.

The April inventory and formal authority instruments then completed further parts of the arrangement.

This is a persuasive explanation. A body responsible for coordinated unique number records should demonstrate that it can operate securely and consistently before it receives unsupervised access and independent standing. Moving live service in stages can be safer than an overnight cutover. Retaining peers during a proving period can identify errors, transfer tacit knowledge and offer a continuity path. The record should not be abused to call AFRINIC a sham, a mailbox or merely a proxy because its decisions were reviewed during transition.

But accepting the strongest accreditation case does not justify enlarging it. ICP-2 recognition could change who the global system treated as the regional registry and what coordinated resources it could receive. It could not create Mauritius corporate personality, which belonged to another legal process. It did not operate the software or handle a February request. It did not transform an allocation record into ownership of a customer network. It did not compel a route announcement or acceptance. It did not delegate police, prosecutorial, judicial or sovereign punitive power.

The evidence therefore defeats two myths at once. The first says recognition was an empty ceremony because the service already ran. That ignores the supervised character of Phase II, the value of coordinated inventory and the institutional consequences that followed final approval. The second says recognition created AFRINIC’s operational authority in one constitutive act. That ignores 46 days of primary service, earlier corporate and policy development, and the later completion of NRO and custodial instruments.

The more accurate conclusion is less theatrical and more useful: operation supplied evidence for recognition, while recognition altered a bounded coordination status.

What remains outside the record

Several unknowns are material enough to shape any future interpretation.

The original transition plan and Phase II acceptance checklist were not located. Without them, it is impossible to compare intended milestones with actual completion at a detailed level. The NRO letter anticipated an 11 April activation as “fully operational” even while saying systems were in production and full operation began on 21 February under second opinion. That wording may distinguish supervised Phase II from recognition-dependent Phase III, or it may reflect planning terminology. No located definition resolves it.

The second-opinion procedure is the largest authority gap. The evidence shows that it existed, but not who invoked it, what decisions required it, whether it bound AFRINIC, how disagreement was settled or how long review took. Request-level records would reveal whether peers mainly assured quality, retained material control, corrected a small number of errors or routinely shaped outcomes. Until then, the mechanism cannot be quantified.

The source pools behind Phase II decisions are also insufficiently documented. AFRINIC was processing requests; incumbent RIRs still had roles; a post-recognition IPv4 and ASN package followed; and the current IPv6 registry shows an earlier AFRINIC line. No pool-by-pool delegation instruments or request-to-pool mapping were located. This blocks a simple answer to who held each piece of resource authority on 21 February.

The APNIC and ARIN sides of the migration lack the detailed object and contract inventories available from RIPE NCC. The actual April authority-transfer letters are absent, as is the Mauritius incorporation certificate. Adoption is also incompletely measured: institutional statements say LIRs and services transferred, while the record lacks a denominator of signed AFRINIC contracts, refusals or exceptions. These are not reasons to doubt the operational handover. They are reasons to stop at its evidentiary edge.

The bounded meaning of 21 February

What happened on 21 February 2005 was both narrower and more consequential than an institutional founding myth. The service moved. Operators were directed to AFRINIC; staff and systems took primary responsibility; records, accounts and DNS administration entered a staged migration. That practical capacity preceded final recognition and became part of the evidence that recognition could responsibly follow.

The event did not collapse every form of control into AFRINIC. Peer review remained. Contractual relationships lagged. Inventory had its own chronology. Corporate personality came from domestic law and has a disputed published date in the available sources. NRO membership and formal authority-transfer instruments came later. Routers remained with operators. Sovereign powers remained with competent public authorities, not a private registry corporation.

The historical achievement is clearer when those limits are stated, not weaker. AFRINIC became a consequential regional ledger and coordinator by demonstrating that it could receive, evaluate, record and publish. Its service mattered because network operators and resource holders relied on coherent records and functioning interfaces. Its recognition mattered because global coordination depends on counterparties accepting one another’s roles and maintaining uniqueness. Neither fact requires pretending that the bookkeeper owns the system described in its books.

The 46-day interval is therefore more than an oddity in a chronology. It is a controlled test of institutional language. Running service, peer supervision, resource custody, accreditation, legal personality and routing operation can occupy different hands at the same time. Good governance begins by naming those hands accurately. Bad governance begins when the word “authority” is allowed to erase their boundaries.