Summary

  • AFRINIC’s current Board register labels the 8 August 2018 ratification of AFPUB-2018-V6-004-DRAFT01 as Resolution 201808.448, while the approved minutes label the same item Resolution 201808.449; the register uses 201808.449 for a different decision.
  • The public record supports four distinct states: proposal and Last Call, Board adoption, incorporation into CPM v1.3 on 1 November, and a late-November implementation report saying that automation of routing and non-disaggregation checks was still pending.
  • The change usefully opened IPv6 PI eligibility to end sites without an IPv4 PI history, retained a /48-or-shorter assignment structure where justified, and created a one-time rectification route for an inadequate initial assignment.
  • Those are legitimate features of a private registry service. They do not give AFRINIC sovereign, legislative, regulatory, police, punitive, confiscatory or public-law adjudicative authority over networks or operators.

Analysis

One act, two numbers, and no magic date

The first fact in this story is a contradiction in AFRINIC’s own documentary chain. The organisation’s current register of 2018 Board resolutions publishes the ratification of the IPv6 PI Update, AFPUB-2018-V6-004-DRAFT01, as Resolution 201808.448. The approved minutes of the Board meeting on 7–8 August print Resolution 201808.449 beside the same proposal. That is not a harmless choice between two equally usable citations: the current register assigns 201808.449 to a separate act authorising the chief executive to sign the host memorandum for AFRINIC-29.

A later entry, Resolution 201809.450, says that the August minutes were approved with amendments, but the examined public record does not disclose what those amendments were. No examined primary source supports the sometimes-repeated label Resolution 201809.448.

An operator therefore cannot responsibly choose one artefact, treat its number as infallible and turn it into a magic effective date. The register is the clearest current index, but it cannot silently erase what the approved minutes actually print. The minutes are evidence of the meeting record, but their number conflicts with a different act in the register. The September approval establishes that amendments existed, not their content. Most importantly, neither number answers the operational question by itself.

The minutes created separate follow-up tasks, the chair’s later announcement described implementation as work still to be done, the manual changed in November, and the late-November meeting report still identified unfinished automation. The honest citation is consequently a described act with a preserved conflict: the Board ratified V6-004 on 8 August 2018; the current register calls it 201808.448; the approved minutes call it 201808.449.

That formulation may look less tidy than an institutional timeline built around a single resolution code. It is more useful. Internet infrastructure depends on version discipline: which text was considered, which corporate body accepted it, which manual carried it, and which service functions could process it. A neat but unsupported identifier disguises gaps in that chain. A careful account exposes them without inflating a records defect into an allegation of deceit. The evidence establishes inconsistency, not motive.

Four states that should not be compressed into one

The proposal began as a specific amendment to a private service manual. AFRINIC published IPv6 PI Update Draft 1 on 28 March 2018 under the exact identifier AFPUB-2018-V6-004-DRAFT01, targeting article 6.8 of the Consolidated Policy Manual. Its immediate problem was practical. Access to provider-independent IPv6 space was entangled with a prior holding of IPv4 PI space, even though an organisation planning an IPv6 deployment might have no such history. The proposal also addressed what should happen when an end site’s original IPv6 PI assignment proved too small for a revised addressing plan.

Those details explain the subject of the later Board act, but they were not yet an adopted or implemented service state merely because a draft had been published.

At AFRINIC-28, the public-policy-meeting record describes discussion on 9 May. The proposal would open the route to applicants without IPv4 PI, retain the HD-ratio material, allow a rectification of an initial assignment, and clarify that staff were not being instructed to monitor routing. One objection to the current version was recorded. The summary reports no movement towards a text change and says that the co-chairs moved the proposal to Last Call. This was the first state: a proposed manual amendment passing through AFRINIC’s own policy-development process.

Last Call was not a Board act, a manual revision or proof that a hostmaster workflow had been changed. It was a process step recorded by the organisation.

The second state arrived at the face-to-face Board meeting in Mauritius on 7–8 August. The Board considered the IPv6 PI Update under agenda item 9.3. Its published operative sentence was bounded: after recitals concerning completion of the policy-development process, a claim of consensus and referral by the Policy Development Working Group chairs, the Board ratified the proposal. The minutes record SO as proposing and SI as seconding the item. They do not print a vote tally or a sentence saying this particular item passed unanimously.

The two preceding policy items do carry explicit unanimous-passage language; that contrast makes it especially important not to borrow their outcome wording for V6-004.

The act was corporate adoption within AFRINIC. It established that the company’s Board accepted the proposal through the organisation’s chosen procedure. It did not legislate for Africa, express the political consent of every network in the service region, or complete the work needed to receive and decide requests under the new clauses. The meeting record itself separates decision from execution. Action Items 201808.22-.23 followed: Action Item 201808.22 assigned the chair to inform the PDWG co-chairs of ratification, while Action Item 201808.23 tasked staff with implementing the ratified policies.

If the vote had been self-executing in every operational sense, the implementation instruction would be difficult to explain. Its presence is direct evidence that another stage remained.

The Board chair’s announcement on 21 August made the separation plainer. It said that V6-004 had been ratified at the 8 August meeting and that staff would proceed with implementation, after which the community would be informed. The announcement supplied no resolution number, so it cannot settle the 448-versus-449 conflict. It does, however, place implementation in the future relative to ratification. The statement is not a technical deployment log, but its tense matters.

It prevents an account in which the Board’s motion instantly transformed every form, assessment routine, database rule and staff practice required to deliver the revised service.

The third state is the versioned manual. CPM v1.3 carries a revision date of 1 November 2018 and records that the IPv6 PI Update changed sections 6.8, 6.8.1, 6.8.2 and 6.8.3 and added section 6.8.4. From that date, a reader could point to a consolidated text in which the adopted eligibility, sizing, subsequent-assignment and rectification clauses appeared. That is a significant improvement over relying on a proposal page or a recollection of meeting consensus. It is also a different proposition from saying that every supporting control had become operational.

A manual can state a rule before all of the organisation’s tooling reliably performs it.

The fourth state appears in the record of AFRINIC-29, held from 26 to 30 November. Both the meeting page and its PDF report the IPv6 PI Update as under implementation. Automation remained pending for policy requirements involving checks that assigned PI space was routed and was not disaggregated. The record does not say exactly which clauses could already be handled manually, which requests were accepted at which moments, or when the automation was ultimately completed. It therefore supports neither the claim that nothing was usable after 1 November nor the claim that the whole regime was finished then.

It supports a narrower, more precise finding: the text had entered CPM v1.3, yet AFRINIC was still reporting part of implementation as incomplete later that month.

These four states are not bureaucratic hairsplitting. They answer different questions. Publication identifies the proposed text. Last Call records an internal process judgement. Board ratification records corporate adoption. Manual incorporation identifies a versioned rule set. Operational reporting describes service readiness. An organisation can move through those states quickly or slowly, but no label causes the evidence from the other states to disappear.

For an operator deciding whether an addressing plan can proceed, the difference between “the Board adopted this” and “the registry can process this consistently” is the difference between institutional expectation and executable service.

What the November manual actually made legible

The benign case for the update begins with a genuine defect in the earlier eligibility route. An end site seeking IPv6 PI space should not have needed an IPv4 PI past merely to demonstrate that it was an end site with a justified IPv6 addressing requirement. Removing that dependency aligned the entry path more closely with the resource being requested. It avoided turning historical access to a scarce older protocol’s provider-independent space into a ticket for participation in the newer one. Within the bounds of a voluntary registry service, that was a sensible correction.

CPM v1.3 described the intended beneficiary as an end site serving a single administrative organisation. The applicant was not to be a Local Internet Registry, was to become an AFRINIC End User Member, had to justify need and plan use within twelve months, and was not to disaggregate the announced PI space. These were service criteria attached to the registry’s assignment decision. Identity, membership status, intended use and an addressing plan are relevant to a registry trying to avoid duplicate assignments and size a block.

Their legitimacy depends on remaining connected to that limited function rather than being treated as a licence to supervise the applicant’s business.

The sizing provisions were similarly concrete. The incorporated text provided for an initial /48, or a shorter prefix when the applicant justified it. In IPv6 notation, “shorter” means a numerically smaller prefix length and thus a larger address block. Subsequent assignments required documentation, and a contiguous extension was preferred where possible. The operational purpose is intelligible.

A default provides predictability; justification allows a plan whose needs exceed that default to be considered; documentation gives the registry a record for later allocation decisions; contiguity can reduce the number of distinct blocks an organisation must administer and announce.

New section 6.8.4 dealt with a common planning failure: the first assignment can turn out to be inadequate. It allowed one rectification of the initial PI assignment, supported by an updated plan. Depending on what was available and justified, that could involve an extension of the same block. Other alternatives contemplated renumbering into a new block within six months or retaining a complementary block. The exact choices mattered because “just obtain another prefix” is not costless advice.

Address plans live inside routers, firewalls, domain-name records, certificates, monitoring systems, access lists, inventory databases, customer configurations and staff procedures. A correction mechanism acknowledged that an initial forecast is not immutable.

The phrase “one rectification” also marked a boundary. The manual was not offering unlimited ad hoc resizing. It was creating a defined chance to repair the initial sizing decision and asking for a revised plan. Whether every detail was optimal is beyond the evidence available here; no request-level data show how many applicants used the route, how quickly cases moved, or what costs were avoided. Still, the structure is recognisable as a private service rule: a default, an evidential exception and a recorded correction pathway.

The manual’s non-disaggregation condition and the later automation report require care. They show that AFRINIC contemplated checks connected to how assigned PI space appeared in routing. They do not prove that the registry was authorised to govern routing at large. At AFRINIC-28, the author’s recorded clarification was that staff were not instructed to monitor routing. Later, AFRINIC-29 described automation of checks concerning whether assigned space was routed and not disaggregated. Those statements can coexist if the later work concerned processing evidence for a defined service condition rather than continuous supervision of the Internet.

The sources do not provide enough detail to describe the exact implementation, so an honest analysis should not invent it.

Nor does a registry assignment guarantee global reachability. A unique number recorded in a registry and a route accepted by networks are related but distinct phenomena. Operators and their upstreams make routing choices; filters and commercial relationships affect propagation; technical conventions influence what is accepted. AFRINIC’s ability to record a PI assignment does not compel every autonomous network to carry it. This distinction keeps the value of the registry clear. The registry helps make assignments unique, attributable and discoverable. It does not author the connectivity that independent networks collectively produce.

Why effective-state clarity carries an operator price

Provider-independent space matters because it can reduce one form of dependency. An organisation using addresses supplied by a connectivity provider may face extensive renumbering when it changes upstream arrangements. A PI assignment can support continuity across such a change, subject to the realities of routing acceptance and operational design. It can give the address plan a life less tightly coupled to one supplier. That is not absolute independence, and it does not erase the costs of operating the prefix, but it can preserve choices that would otherwise be expensive to exercise.

Initial sizing is where an apparently administrative rule can meet physical work. If an organisation builds around a /48 and later concludes that the original plan was inadequate, the consequences are not confined to a form sent to a registry. Engineers may need to redesign subnet boundaries, renumber hosts and interfaces, revise firewall objects, update name-service records, replace hard-coded allowlists, alter monitoring targets, coordinate customers and suppliers, and manage a transition in which old and new values coexist. Certificates and security controls may embed addresses or depend on systems that do.

Support teams must explain failures during the change. A larger contiguous block, if justified early, can avoid some of that disruption; a rectification route can reduce the damage when the forecast was wrong.

Multiple-prefix operation has its own trade-offs. Retaining a complementary block may avoid immediate renumbering but adds another object to routing policy, security policy, asset tracking and incident response. Moving into a new block can create a cleaner long-term plan while concentrating transition cost into a limited period. Extending the same block can be operationally attractive but depends on adjacent availability and registry judgement. CPM v1.3’s alternatives did not make those costs disappear.

They made it possible to choose among them within the service record rather than pretend that an initial assignment could never need correction.

Eligibility uncertainty also has a price. Network plans interact with equipment orders, staff scheduling, software work, upstream negotiations and customer commitments. If an applicant knows only that a proposal reached Last Call, it cannot assume the Board will adopt it. If it knows only that the Board ratified the proposal, it still needs to know when the manual reflects the decision and whether requests can be processed. If the manual has changed but relevant checks are still being automated, the applicant needs an accurate statement of what can be done manually and what remains unavailable.

Each ambiguous hand-off can force a delay, a contingency plan or duplicated work.

This is why the numbering conflict deserves attention without melodrama. Operators rarely build a deployment around a resolution code alone. Yet identifiers connect artefacts. They allow a legal team, engineer, registry analyst or auditor to establish that the Board item, action instruction, manual redline and implementation notice concern the same change. When two primary artefacts attach different numbers to one act—and one of those numbers identifies another act elsewhere—the cost is extra interpretation. The user must reconstruct the chain instead of following it.

The cost need not be catastrophic to be real. A records system is valuable precisely because it reduces the need for every participant to perform historical detective work. Accurate identifiers lower search and verification costs. Dated revisions reduce disagreement over which text applied. Explicit readiness notices let operators distinguish a published promise from an available workflow. When those elements are absent or inconsistent, prudent users must add their own checks. The institution has shifted a small but unnecessary part of its coordination burden back to them.

The later NRS operator guidance illustrates the kind of practical workflow an applicant may encounter: a distinction between LIR and PI eligibility, organisation documents, network diagrams, deployment plans, hostmaster evaluation, payment and a registry entry. That guidance was published in 2025 and cannot be used as evidence of the wording or readiness of the 2018 regime. Its value here is contextual. It reminds us that an “assignment policy” becomes real for an operator through a sequence of documents, assessments, fees and records. A Board resolution is one upstream artefact in that sequence, not the sequence itself.

The 2018 annual report likewise provides confirmation at the correct level and no more. It associates the IPv6 PI Update with August ratification and with CPM sections 6.8–6.8.4. It does not cure the inconsistent numbering, disclose the September amendments or establish the date on which pending automation was completed. Annual summaries compress. An effective-state analysis has to decompress the record and retain the qualifications that a year-end overview omits.

The strongest benign reading

The most persuasive defence of AFRINIC’s handling starts from the usefulness of structured change. A private registry needs a way to revise its service rules. Here a proposal identified a real eligibility problem and a foreseeable sizing problem. The public process recorded discussion and an objection. Co-chairs advanced the text through Last Call and referred the outcome. The Board used a corporate adoption step. Staff received an implementation task. The organisation then incorporated the clauses into a dated manual and worked on automating checks associated with them. That is a more legible method than undocumented discretion.

On substance, the change widened a continuity option. It allowed an end site without the historical accident of IPv4 PI space to seek IPv6 PI space on its own merits. It retained a minimum assignment structure, demanded justification for a larger block and created a single rectification route. Those are not inherently instruments of control. Properly confined, they can make a non-duplicative registry service fairer and more predictable. An applicant can know the baseline, explain an exception and return once if its initial plan needs correction.

Some verification is not merely permissible but necessary to the service. A registry that does not authenticate the organisation requesting a block risks error or fraud. One that does not check whether a requested number collides with an existing assignment fails at uniqueness. One that never asks about size cannot rationally choose among available blocks. One that keeps no reliable contact and assignment records deprives operators of the coordination benefit for which the registry exists. Membership can define the contractual relationship through which the service is delivered.

These are bounded administrative inputs, not evidence of public jurisdiction.

Automation can also improve fairness if it faithfully implements disclosed criteria. A repeatable check may reduce arbitrary handling, shorten staff work and produce a record that similar applications received similar treatment. The late-November report of unfinished automation is therefore not necessarily evidence of an improper delay. The record gives no basis for that inference. It is evidence that work remained. The distinction matters: transparency about incomplete implementation is useful, even when a fuller readiness statement would have been better.

The case for restraint should therefore not be mistaken for an attack on IPv6 PI, portability, routing hygiene or technical validation. The update’s operational benefits can be accepted without converting AFRINIC’s process vocabulary into constitutional language. Indeed, giving the bounded act full credit makes the authority question easier to see. The registry did something worthwhile when it corrected an entry criterion and published a versioned route for assignments and rectification. It did not need sovereign power to do so.

A private service rule is not public law

AFRINIC is a Mauritius-incorporated private technical bookkeeper and uniqueness coordinator. Its legitimate role is substantial but narrow. It can maintain a ledger of assignments, authenticate parties using its service, avoid allocating the same resource twice, keep contact information, evaluate a documented technical need for an assignment and publish the state of its own manual. It can set contractual terms for voluntary participation in that service, subject to the law that actually governs it.

Nothing in the V6-004 record creates sovereign, legislative, regulatory, police, prosecutorial, punitive, confiscatory or public-law adjudicative authority.

The word “ratified” can encourage conceptual slippage because it sounds weightier than “the Board approved a change to the company’s service manual”. The operative reality is the latter. The Board could bind AFRINIC’s corporate machinery to implement the accepted proposal. It could not, by giving the act a resolution number, make rules for every African network regardless of participation. The Policy Development Working Group chairs could report a process outcome. They could not transform meeting attendance, mailing-list participation or silence by absent operators into a continent’s political consent.

The same applies to “consensus”. In this record, consensus is an institutional claim used within AFRINIC’s procedure. It helps explain why the proposal was referred and why the Board considered ratification. It does not prove representation of every affected operator, still less grant coercive power. A public-policy meeting can be a valuable forum for technical coordination while remaining a gathering organised by a private institution. Its conclusions may guide that institution’s services. They are not legislation.

This boundary is not semantic pedantry. It determines what follows from non-compliance. If an applicant does not satisfy the documented criteria for a particular voluntary assignment service, AFRINIC may decide that the application has not established entitlement under that service. That does not make the applicant a lawbreaker. It does not authorise punishment of unrelated business conduct, seizure of property, policing of all routes or adjudication of public rights. The registry’s conclusion must remain tied to the narrow transaction it is equipped to perform.

Technical recordkeeping must likewise be separated from network control. A registry can ask for an addressing plan to size an assignment. That does not give it a continuing mandate to design the network. It can record a condition concerning the prefix requested through its service. That does not make every independent routing choice an offence against a regulator. It can maintain a contact record that assists coordination. That does not turn the ledger into the source of the operator’s technical existence.

The bookkeeper metaphor is exact. A good bookkeeper records a transaction faithfully, preserves provenance, reconciles inconsistencies and makes the account intelligible to the people who rely on it. The bookkeeper does not create the underlying economic reality merely by entering it. In Internet numbering, the registry’s record is valuable because independent systems can refer to a common, non-duplicative assignment. The usefulness of that reference does not make the registry sovereign over the networks that put addresses into operation.

This is the controlling insight of Heng Lu’s Why RIRs Do Not Have Authority, The Policy Mirror and On When the Bookkeeper Auditions for Olympus. The common layer should be thin: uniqueness, accurate records, contactability and operational continuity. When institutional language presents stewardship, community or policy as if those words themselves created jurisdiction, it confuses a coordination service with government. Why IPv6 Was Pushed, and Who It Actually Serves adds the operator-economic discipline: deployment costs are paid by the organisations doing the work. Administrative promotion of IPv6 does not transfer those costs to the registry or create a mandate to command the investment.

LARUS’s operator analysis of how registry governance decisions can reach infrastructure reinforces the transmission mechanism without proving any additional fact about the 2018 vote. Unclear service state affects planning because operators must act on the registry’s outputs. BTW’s analysis of transition architecture similarly distinguishes the value of durable records from discretionary gatekeeper power. These sources are useful for interpreting institutional scope. They do not settle the resolution number, and they should not be made to do so.

Official AFRINIC records establish what AFRINIC recorded and did; the operator and institutional analyses explain why the limits of those acts matter.

A thin authority boundary also protects the best parts of the update. When identity checks, documented need and assignment sizing are presented as tools for accurate, unique service delivery, they can be judged against that purpose. Are the criteria necessary? Are they applied consistently? Is the text versioned? Can an error be corrected? When the same checks are spoken of as broad governance of a “community”, accountability becomes harder because the claimed mission has no natural edge. Restraint makes performance measurable.

What the evidence does not permit

The public record is unusually instructive because it contains both concrete facts and visible holes. It proves the proposal identifier, submission date and manual target. It proves the Board considered the item and ratified it on 8 August. It proves two official artefacts print incompatible resolution numbers. It proves that notification and implementation were separately assigned. It proves the November manual carried the revised sections. It proves that automation was still described as pending late that month. Each statement has a documentary anchor.

It does not prove why the number differs. “Typographical error” may be plausible, but plausibility is not a disclosed correction. It does not reveal the amendments attached to the September approval of the August minutes. It does not establish a V6-004 vote tally or unanimity. It does not show that all functionality was unavailable until automation finished, nor which clauses staff might have processed manually. It does not give a completion date for automation.

Nor does it quantify outcomes. There are no request-level figures here for applications, approvals, refusals, processing times, fees, rectifications, avoided renumbering or routing effects under the 2018 change. It would therefore be unsound to claim that the update produced a measured increase in IPv6 adoption, saved a stated sum, harmed a particular applicant or caused a documented routing event. The economic mechanisms are credible because they describe the work associated with address changes; their realised scale in this programme is unknown.

Later clarification must remain later. Changes and explanatory material from 2020 onwards may describe a subsequent regime, but they cannot be read backwards as proof of what sections 6.8–6.8.4 meant or how they were implemented in November 2018. Version control exists to prevent precisely that form of temporal blending. The relevant text is CPM v1.3, qualified by the contemporaneous implementation report.

The lack of evidence for misconduct is equally important. A conflicting identifier and incomplete automation justify criticism of documentation and readiness transparency. They do not justify an allegation of deliberate falsification, concealed motive or punitive intent. Institutional analysis becomes stronger when it refuses both deference and speculation. The register does not become correct merely because it is official; the conflict does not become sinister merely because it exists.

This discipline also prevents the story from expanding into unrelated disputes. The merits of other 2018 IPv6 proposals are not necessary to understand V6-004’s Board act. Neither are later transfer questions, current governance controversies or general histories of regional registries. The narrow chain already supplies a complete analytical problem: a useful service amendment, a conflicted instrument label, a staged implementation and a temptation to confuse private procedure with public authority.

A better effective-state receipt

The restrained alternative is not to abandon the policy-development process or remove all verification. It is to make the service state legible with fewer interpretive leaps. For this change, one canonical resolution record should have linked directly to the approved minutes and used the same identifier in both places. If a number was corrected, the record should have preserved the original, identified the correction and stated when it was made. Resolution 201809.450’s unspecified amendments should have been published or linked so readers could determine whether they addressed the conflict.

The corporate act should also have travelled with a signed or otherwise authenticated redline. The proposal archive identifies the draft, and CPM v1.3 shows the consolidated result, but an explicit redline would connect adopted clauses to the resulting manual text. That matters when several changes are processed near one another. It lets operators check that the Board act and the manual revision are the same instrument rather than infer the connection from a revision note.

Most valuable would have been a clause-by-clause readiness notice. It could have stated the date on which the manual text governed AFRINIC’s internal handling, which requests staff could process manually, which checks had been automated and which remained pending. If routing and non-disaggregation checks delayed only one workflow, the notice could have isolated that limitation rather than leaving “under implementation” to cover the entire update. When automation completed, the same notice could have recorded the date and version.

The decision receipt for an applicant could then remain narrow: verified identity; a non-duplicative registry check; End User Member status; the documented need and intended twelve-month use relevant to sizing; the assignment issued; and, if invoked, the selected rectification route. Such a receipt would support consistency and auditability without pretending to decide the applicant’s wider business, architecture or legal rights. It would show what the registry knew, what it checked and what service decision followed.

Routing would stay where routing belongs. The registry could disclose the assignment and the service condition it applied, but independent networks and counterparties would continue to make acceptance decisions. Business investment would stay with operators. Punishment, confiscation and adjudication would stay with institutions that possess lawful public authority, not be inferred from an address ledger. The result would preserve the operational benefit of a common record while preventing the record from becoming a claim of ownership over the system it describes.

This alternative is not utopian. It asks for ordinary properties of reliable technical administration: a stable identifier, a traceable text, a dated state transition, explicit readiness and a limited decision record. Those controls are especially important when the service affects long-lived infrastructure. They reduce disputes not by asserting institutional supremacy but by making the institution’s own actions easier to verify.

The real lesson of Resolution 201808.448/201808.449

The 8 August act did something concrete and useful. It accepted AFPUB-2018-V6-004-DRAFT01 within AFRINIC, moving an IPv6 PI route beyond an obsolete IPv4 dependency and towards a clearer assignment and rectification framework. CPM v1.3 subsequently gave that framework a dated, consolidated form in sections 6.8–6.8.4. The value should not be understated: continuity options, predictable sizing and a way to repair an initial plan are material to operators.

But the act did not do everything later shorthand may place upon it. It did not settle its own public number consistently. It did not prove unanimity. It did not execute Action Item 201808.23 at the moment of the vote. It did not make the 1 November manual revision proof that every automated check was ready. And it did not turn a private registry into a public regulator.

The best reading is neither hostile nor credulous. AFRINIC used its corporate machinery to improve a bounded service, recorded the result imperfectly, and continued implementing it after the manual changed. Operators should recognise the benefit, insist on the qualifications and refuse the jurisdictional leap. A registry earns trust by authenticating, assigning uniquely, keeping accurate records and reporting its effective state. The less it asks a resolution number to signify beyond those functions, the more dependable that number becomes.