Cloud Innovation's Contract and the Limits of Retrospective Interpretation

Summary

• The decisive instrument was the agreement Cloud Innovation signed in July 2013, together with the allocation applications, adopted policies and any later terms validly incorporated through renewal; a publicly available 2017 agreement could illuminate later wording but could not silently stand in for the executed 2013 text.
• The record required separate findings on eligibility, stated need, purpose, cooperation, accuracy of information, transfer or leasing, out-of-region use and post-allocation change. Combining those questions into a general claim that addresses were not serving Africa risked replacing clauses with institutional intuition.
• AFRINIC's own archive records that a 2014-2015 proposal described existing policy as not explicitly allowing or disallowing out-of-region use and was withdrawn in 2016. That silence did not decide every contractual issue, but it made retrospective reliance on a supposedly obvious geographic rule unsafe.
• Cloud Innovation's 2020 response disputed AFRINIC's routing methodology and legal interpretation. Those were party submissions, not established facts; they nevertheless required a particularised answer based on reproducible measurements and the text governing each allocation.
• The July 2021 injunction ruling turned on preliminary representation issues, and the appellate court later expressly declined to decide the merits. Procedural outcomes should not be repackaged as judicial confirmation of the alleged contractual breaches.

There were four relevant clocks, not one timeless bargain

The dispute between AFRINIC and Cloud Innovation was often presented as a test of stewardship: did a regional registry have the power to ensure that addresses issued for African need served Africa? That framing is politically potent and legally incomplete. Contractual compliance is determined through instruments and facts attached to dates. The dates do not disappear because the resource is scarce or the later policy argument seems compelling.

At least four clocks were running. The first started when Cloud Innovation signed a service agreement on 23 July 2013, a date the company recorded in its July 2020 response. The second ran with each application and allocation, because the evidence of need, purpose and planned network use could differ from block to block. The third ran with changes to AFRINIC's agreement and community policies, including renewal terms. The fourth ran with actual use, customer assignments, communications and AFRINIC's later review.

A serious breach analysis aligns those clocks. It asks what promise bound the member when a particular act occurred, whether a later renewal validly changed the promise, what facts were represented for a specific allocation, and what the member actually did. It does not begin with a present rule and project it backwards over every earlier event.

This temporal discipline protects both parties. Cloud Innovation could not freeze its duties in 2013 if it later accepted a renewal that incorporated valid amendments or future adopted policies. AFRINIC could not cite the most convenient later text without proving that it governed the member and the relevant conduct. The answer might vary by year, clause and address block. That is not pedantry. It is what makes enforcement predictable rather than discretionary.

The public controversy compressed these clocks into a moral story. AFRINIC used a land-manager analogy in which a plot granted for an African hotel became a shopping centre elsewhere. Cloud Innovation argued that policy did not prohibit its model and challenged AFRINIC's technical inferences. Neither analogy nor denial is the contract. The institution needed the signed instruments, application evidence, amendment notices, renewal record and measured use.

The executed 2013 agreement was the indispensable first document

Public materials establish that an agreement was signed in July 2013, but the executed copy and complete allocation files are not reproduced in the sources used for this analysis. That absence sets a hard limit. One cannot responsibly attribute every clause of the agreement AFRINIC published in November 2017 to the contract signed four years earlier.

The 2017 change paper itself proves why. AFRINIC described proposed revisions from a January 2016 version. It changed terminology, definitions and transfer treatment, edited references, and clarified what would happen in case of breach. It also defined adopted policy to include future policy and revised language connecting resources to stated purpose. A redline exists because text changed. The later clean agreement is evidence of the later terms, not conclusive evidence of the earlier ones.

The first step in adjudication should therefore have been documentary authentication. Produce the agreement bearing both parties' signatures. Identify schedules, online terms or policies incorporated by reference. Establish its duration and renewal mechanism. Produce every amendment notice and evidence of delivery. Show the terms prevailing at each renewal, including the renewal effective on 1 January 2021 that the July judgment noted.

Then identify whether the agreement allowed unilateral amendments, future policies or annual renewal on prevailing terms. The 2017 published agreement said amendments would take effect after posting and mailing-list notice, and that refusal would stop services. It also defined adopted policy to include policy adopted during the term. Whether materially similar mechanisms existed in 2013, and whether the required notice occurred, must be proved from the governing documents rather than presumed.

This approach does not create an escape through archival uncertainty. A party that signed a clear cooperation, purpose or accuracy obligation remains bound to it. The point is that clarity must be shown from the contract, not manufactured by quoting a later edition. AFRINIC was the institutional drafter and record keeper; preserving executed versions and notices was part of credible enforcement.

Allocation purpose and regional-use policy were related but not identical

AFRINIC's public case rested heavily on purpose. Its later registration agreement says an applicant has an exclusive right of use within the need justified in its application and for no other purpose. Its litigation page portrayed Cloud Innovation as having obtained space for African services and then using only a tiny portion in Africa.

If comparable purpose language governed the relevant allocation, and if an application made a clear factual commitment that the addresses would support a specified African deployment, material departure could be a contractual issue even without a general policy banning all out-of-region use. A party cannot necessarily avoid a specific representation by pointing to silence in a broader policy.

The converse is equally important. A statement of eligibility as a company based in the service region is not automatically a promise that every customer, server or route will remain physically in-region. A general description of providing cloud, hosting or address-management services may permit several architectures. A need justification expressed through projected customers may change over time. The exact words and context determine the commitment.

This is where the public land analogy becomes dangerous. Land is territorially fixed and generally used through possession. Internet addresses are globally routable identifiers administered through a regional allocation system. A company based in the region can support customers elsewhere; a route originated by an ASN registered elsewhere can carry regional use; and an aggregate may conceal more-specific announcements. The analogy may communicate AFRINIC's view of purpose, but it cannot supply the missing technical or contractual elements.

The decision-maker needed a block-level table. For each allocation: application date, agreement version, policy version, stated service, forecast need, represented geography, approved quantity, later customer assignments, observed routing, member explanations and alleged breach. Without that table, “mostly outside Africa” could blend distinct allocations and distinct promises into one conclusion.

The withdrawn out-of-region proposal is evidence of genuine policy ambiguity

AFRINIC's policy archive contains an unusually revealing document. A proposal initiated in July 2014 sought to govern out-of-region use. Its problem statement said existing AFRINIC policies did not explicitly allow or disallow such use and warned that the silence left staff to decide arbitrarily. The proposal contemplated permitting bounded external use after regional use, went through revisions and was withdrawn in February 2016.

This history does not establish that every out-of-region arrangement was lawful. A draft never adopted is not policy. Existing contractual commitments, eligibility rules, need requirements and the soft-landing restriction for particular space could still constrain conduct. The proposal's author did not adjudicate Cloud Innovation's contract.

But the archive defeats a simplistic claim that one unambiguous regional-use rule had always covered every address and every form of external use. The community was debating the gap precisely because entities considered the answer insufficiently explicit. One revision removed retrospective application language. Another sought clearer criteria for determining geographic use and reduced staff discretion. Those changes reveal concern about both time and proof.

Policy silence is not a licence; it is a reason for contractual precision. AFRINIC could identify a false application statement, breach of a specific purpose clause, failure to cooperate, unsupported change, impermissible transfer or a rule applying to final-/8 space. What it should not do is turn the policy outcome it later preferred into an unwritten term of earlier allocations.

There is also a constitutional dimension within the registry model. AFRINIC describes resource policy as developed through an open, bottom-up process. If staff can impose a major geographic rule through case-by-case interpretation after a proposal on that subject failed to become policy, the distinction between administration and policy formation weakens. Enforcement must remain strong, but it must enforce adopted commitments rather than create policy through sanction.

The contract could still carry future policy, but incorporation had to be proved

Temporal interpretation is not the same as insisting that only the 2013 rulebook mattered forever. Long-running service relationships often anticipate change. AFRINIC needed to respond to exhaustion, transfers, security and new community policies. A member renewing annually may accept terms that incorporate later adopted policy.

The 2017 agreement published by AFRINIC expressly defined adopted policy to include policy adopted during the term. It also said annual renewal would occur on terms prevailing for the additional year when specified conditions were met. If those provisions validly governed Cloud Innovation's later renewals, a new policy could bind future conduct without being retroactive.

Three questions follow. Was the later policy adopted through the required community process? Did the agreement clearly incorporate it? Did the alleged conduct occur after the policy became effective or continue into that period? A continuing use can be regulated prospectively even if it began lawfully. The remedy should address post-effective conduct, not relabel the earlier period.

Notice remains important. The published agreement contemplated posting and mailing-list communication for amendments. AFRINIC should show the version sent, date, channel and renewal acceptance. Merely maintaining a current form on a website does not prove which text a member accepted years before. Conversely, payment and continued service after clear notice may have contractual significance depending on the agreement and law; a member cannot necessarily ignore communicated terms while taking the benefit of renewal.

This is why the 31 December 2020 renewal communication matters. AFRINIC's chronology says it informed Cloud Innovation that renewal for 2021 was subject to and without prejudice to the continuing review. That statement preserved a dispute; it did not itself identify which substantive terms governed or prove the breach. A reservation is not a substitute for a clause.

“Use”, “assignment”, “sub-allocation”, “lease” and “transfer” required definitions

The controversy regularly shifted among words that may describe different conduct. Use can mean routing an address, assigning it to a device, enabling a customer, maintaining the registration or exercising contractual control. An assignment can grant a customer operational use while the member retains the allocation. A transfer changes the recognised resource right from source to recipient under policy. Leasing may describe a commercial arrangement without identifying what authority or control actually moved.

The 2017 RSA revision was driven in part by an adopted intra-region transfer policy. AFRINIC's change paper said the existing RSA did not permit transfers and required revision to implement the new policy. That history demonstrates that transfer was a formal category, not a synonym for every customer assignment.

To prove impermissible transfer, AFRINIC needed facts showing that Cloud Innovation relinquished the relevant right or that another entity effectively exercised the authority reserved to the member. Customer contracts, registry contacts, payment structure, duration, control over assignments, routing-security authority and ability to reclaim customer use could all matter. A public route from a customer's network would not alone prove transfer.

To prove that leasing breached purpose or policy, AFRINIC needed the governing prohibition or a demonstration that the arrangement contradicted the represented need. A business label could be evidence, but substance should control. A managed hosting arrangement and a bare monetisation of addresses may be sold under similar language while allocating control differently.

Cloud Innovation, for its part, could not answer every concern by saying policy was silent on geography. It still had to show that customer assignments reflected genuine need, that data supplied to AFRINIC were accurate, that it retained any authority the agreement required, and that changes were communicated as the contract demanded. Temporal fairness is bilateral: it protects the member from invented duties and the registry from evasive relabelling of clear ones.

Technical proof had to match the proposition

Cloud Innovation's July 2020 response challenged AFRINIC's apparent use of ASN registration location, routing views and aggregate prefixes. It argued that an ASN's registered country did not reliably establish where traffic or infrastructure was located, and that more-specific routes and WHOIS assignments mattered. These are submissions by the affected party, not neutral findings. They nevertheless identify testable methodological questions.

Geographic use cannot be established by one convenient proxy. ASN registration reflects an administrative record, not necessarily physical router location or customer location. BGP collectors observe paths from particular vantage points. Geolocation databases infer location and may disagree. WHOIS records reflect registered assignments, but accuracy is itself a compliance issue. Traceroute can show path clues while being affected by addressing, filtering and network design.

The evidence should therefore be triangulated and time-specific. For each sampled prefix, AFRINIC could preserve BGP observations from multiple collectors, more-specific announcements, reverse DNS, customer assignment records, infrastructure invoices, service descriptions, traffic evidence where lawful, and the member's explanation. The result should distinguish registered holder, announcing network, equipment location, customer location and service beneficiary.

The proposition must also be defined. If the clause requires the resource to support connectivity back to Africa, evidence of an African service dependency may matter even when equipment is abroad. If the application promised hosting in a named African facility, facility evidence matters more. If the issue is customer need, geography may be secondary. Measurement cannot rescue an undefined obligation.

AFRINIC's evidence should be reproducible enough for challenge without exposing customer secrets or security details. The notice can identify the prefixes, dates, data types and inference. Cloud Innovation can provide contradictory measurements. An independent expert can test both. Confidence should be stated, especially when allocation-wide conclusions are extrapolated from a sample.

Cooperation was a separate and potentially clearer obligation

The later published RSA required members to provide information, assistance and cooperation reasonably requested for services or utilisation review. Failure could affect service and membership. Similar wording may have existed earlier, but the executed text must confirm it.

Cooperation can be easier to adjudicate than contested geography. AFRINIC should identify the request, its relevance, deadline, information supplied, omissions and follow-up. The member should explain burden, confidentiality, impossibility or why the request was not reasonably connected to the review. A refusal is not proved by disagreement with AFRINIC's conclusion; a detailed rebuttal may be cooperation even if AFRINIC finds it unpersuasive.

Cloud Innovation responded in July 2020 and, according to AFRINIC's chronology, again engaged in the sequence leading to a March 2021 show-cause letter. The existence of responses does not prove completeness. AFRINIC's later claim that breaches were not remedied does not prove non-cooperation unless the outstanding requirement is particularised.

The registry should have issued an evidence schedule. For each item: request date, clause, purpose, exact material sought, response, deficiency, cure required and consequence. This would separate a failure to answer from a substantive dispute about what the answer showed. It would also prevent moving targets, where each response generates a new unstated standard.

Reasonableness matters because the registry holds asymmetric power. It can demand evidence necessary to protect scarce resources, but customer-level disclosure may involve privacy, security and commercial sensitivity. Confidential submission, sampling and independent verification can satisfy oversight without public exposure. A member should not be able to invoke confidentiality as a blanket shield, and AFRINIC should not treat reluctance to publish sensitive records as defiance.

The March 2021 notice needed a breach ledger, not a conclusion

AFRINIC's chronology says its 10 March 2021 letter set out breaches and gave Cloud Innovation thirty days to remedy or show cause why the agreement should not be terminated. The later published RSA includes a thirty-day show-cause structure. Procedurally, notice and an opportunity to cure are important.

Substantive fairness depended on the notice's granularity. Each alleged breach should identify the relevant block, act, date, contract version, incorporated policy, application representation, supporting evidence, requested cure and continuity consequence. “Out-of-region use” is not sufficiently precise if the agreement permits some external operation or if different blocks were issued under different phases. “Leasing” is not sufficiently precise without the prohibited conduct and evidence of control.

The cure must be possible. If the breach is inaccurate information, correction and supporting documents may cure it. If it is unauthorised customer use, transition or reassignment may be required. If AFRINIC believes the original application was false, the issue may not be curable merely by changing present routing. The notice should say which theory applies.

The decision after thirty days should answer the response. Cloud Innovation's technical and legal objections could be rejected, but not ignored. A reasoned decision would explain why the executed agreement covered the conduct, why the measurement method was reliable, which facts were found and why the proposed cure failed. Institutional conviction is not a reason.

Termination is especially serious because the published later agreement linked it to immediate revocation of resources and cessation of services. A block-level remedy may be more proportionate when only some allocations or acts are proved defective. If the agreement mandates entity-level termination, the decision should still assess operational transition and affected customers.

The July 2021 judgment did not decide the contract

Public accounts often turned the set-aside of Cloud Innovation's first injunction into a merits signal. The judgment itself is narrower. The Judge in Chambers focused on authority and representation in affidavits and powers of attorney. The issues listed concerned who could represent the foreign company and provide evidence in the proceeding. The ruling did not conduct the block-by-block contractual analysis outlined here.

The appellate judgment delivered in February 2022 is clearer still. It expressly said the court did not propose to deal with the merits of the remaining appeal grounds. It set the appeal aside after considering overlapping applications and the interim relief already obtained. Whatever criticism it made of the litigation sequence, it did not decide whether Cloud Innovation had breached the 2013 agreement through out-of-region use, leasing or another act.

This distinction is not pro-Cloud Innovation. A procedural ruling can be entirely correct while leaving a weak substantive defence untouched. Nor is it anti-AFRINIC. The registry did not need an injunction judgment to believe its evidence supported enforcement. But public legitimacy requires accuracy about judicial authority.

Saying that AFRINIC “won” a procedural application may be colloquially understandable. Saying the court confirmed the contractual allegations would be inaccurate on the cited record. The unresolved merits needed adjudication through the main contractual case, an agreed appeal mechanism or another competent process.

The misuse of procedural outcomes has a governance cost. Members may believe the hard evidential questions have been settled and stop demanding the contract, allocation files and measurements. Opponents may treat every court setback as proof of institutional bad faith. Precision lowers the temperature and returns attention to the documents that can actually decide the dispute.

Retrospective interpretation is tempting when policy failed to keep pace

IPv4 scarcity gave AFRINIC a legitimate reason to scrutinise large allocations. A regional registry cannot remain indifferent if addresses justified by African need are monetised under materially different arrangements. Members that complied with strict need rules would reasonably expect equal enforcement.

The temptation is to repair an earlier policy gap through interpretation. If the community would probably reject a practice today, administrators may describe it as always prohibited. That route is fast and corrosive. It punishes a party without the notice that policy formation is meant to provide, and it lets staff achieve through enforcement what the community did not adopt through policy.

The sounder response has two tracks. Enforce clear old promises against provable old and continuing facts. At the same time, adopt prospective policy that defines geography, leasing, customer assignment, measurement, reporting and transition. The first protects the bargain; the second closes the gap.

Prospective policy can still affect existing allocations where the agreement validly incorporates future policy and the rule lawfully applies to continuing conduct. It should provide an implementation date and reasonable transition. That is not weakness. It makes compliance observable and uniform.

AFRINIC's withdrawn out-of-region proposal showed that community entities understood the need for explicit criteria. Its history included debate over past application and staff discretion. Those concerns should have informed the 2020-2021 enforcement posture. The more contested the policy baseline, the more exact the contractual case needed to be.

Equal treatment required a portfolio review

Selective enforcement can be legitimate when risk differs, but it should be explained. Cloud Innovation held a large address portfolio and publicly described a global customer model. Size and unusual use could justify priority review. They could not justify a unique rule.

AFRINIC should have stated the selection criteria and tested comparable members. How many large allocations had material out-of-region announcements? Which members assigned space to global customers? What evidence distinguished ordinary multinational network support from prohibited disposition? Were the same information requests, measures and cure periods applied?

A portfolio review would clarify interpretation. If many members used some resources abroad with AFRINIC's knowledge, that course of performance could be relevant to ambiguity, though it would not override clear text. If Cloud Innovation's arrangements transferred control in a way others did not, the evidence would sharpen the case. If the issue arose from scale rather than kind, proportional requirements could be stated transparently.

Equal treatment also protects AFRINIC. A published risk method rebuts claims that enforcement was personal or opportunistic. It lets members see that scrutiny follows block size, unexplained change, inconsistent need evidence or transfer signals. Decisions can remain confidential while aggregate criteria are public.

The portfolio approach should use a common breach ledger and independent review. A member cannot point to another possible violation as a complete defence, but persistent inconsistency undermines legitimacy. The institution should correct both rather than make one controversial case carry an unwritten regional standard.

Continuity belonged in the remedy, not the merits

Cloud Innovation and AFRINIC each invoked consequences for Internet users. Operational dependency matters, but it should not decide whether a clause was breached. A large customer base does not make an invalid arrangement valid. Regional importance does not make an ambiguous duty clear.

Continuity should shape interim and final remedies after the evidential analysis. During review, AFRINIC could prevent new allocations, transfers or material changes while preserving existing services. If breach were proved, it could require a staged customer transition, prefix-specific correction and notice before revocation. Security and abuse response should remain available under supervised conditions.

The remedy should distinguish Cloud Innovation's membership from downstream operators. Customers may have no knowledge of the application representations or contract dispute. They need accurate notice and time to renumber. Claims of enormous user impact should be evidenced, not accepted as a veto. AFRINIC should ask for customer counts, critical services, migration plans and technical dependencies under confidentiality.

A continuity schedule would also discipline AFRINIC's ninety-day grace period. The institution described it as exceptional and humanitarian. A stronger basis would be a published rule tied to measured migration difficulty, security risk and third-party innocence. Similar cases would then receive similar protection.

Keeping continuity separate from merits improves both. The contract is interpreted without fear that a correct finding automatically causes an outage. The transition is designed without pretending that operational dependence cures breach. Courts can preserve the status quo narrowly while the underlying rights are determined.

The allocation file was part of the bargain, but only if it was made definite

AFRINIC's strongest theory may not have required a universal geographic ban. If Cloud Innovation obtained a particular block by presenting a defined African deployment and the agreement limited use to the justified need or requested purpose, the application file could supply factual content to the contractual promise. That theory is narrower than saying all external use was forbidden, and potentially more defensible.

It still required care. Resource applications often contain forecasts, topology descriptions, customer estimates and staged utilisation plans. A forecast that later proves optimistic is not automatically a false representation. A network design may change while continuing to serve the same commercial need. A request may describe initial use without promising that every future assignment will follow the same architecture. The decision-maker must distinguish description, estimate, condition and warranty.

Definiteness begins with the form. Which questions did AFRINIC ask? Did the answer identify named facilities, customer classes, countries, service type or deployment dates? Did AFRINIC request clarification? Did the approval repeat any condition? Was the allocation released in tranches after utilisation checks? If the institution relied on a particular representation, the contemporaneous approval should show that reliance.

The file must also be complete. A spreadsheet extracted years later can omit explanatory tickets, calls or attachments. The member may have disclosed international customers or remote infrastructure in follow-up correspondence. AFRINIC may have approved a revised plan. Conversely, the member may have made a precise commitment in a document not visible in public debate. Only the retained sequence can show the bargain.

Changes after allocation require the same discipline. The later published agreement obliged applicants to update information that changed or became outdated and to answer enquiries. If materially similar duties governed Cloud Innovation, the member could not rely indefinitely on an old plan while operating a different model. But AFRINIC should identify which datum changed, when notice was required, what notice it received and why a WHOIS update did or did not satisfy the duty.

The remedy should follow the materiality of the representation. An inaccurate contact field calls for correction. A delayed deployment may call for revised milestones. A deliberate need statement used to obtain a block that would not otherwise have been allocated can support stronger action. Combining all inaccuracies under one termination theory would obscure why the bargain failed.

This application-centred approach also answers the policy concern prospectively. AFRINIC can make future forms explicit: describe in-region need, expected out-of-region support, customer assignment model, control retained by the member, permitted changes and mandatory notification. Applicants then know what is decisive, and staff have a stable record against which to review later use.

Interpretation needed a hierarchy of authority

The dispute drew on the agreement, bylaws, resource policy, application materials, board decisions, staff practice and AFRINIC's general mission. Those sources did not have equal force or perform the same function. A defensible decision should state the hierarchy before selecting an answer.

The signed agreement governs the bilateral service relationship, subject to applicable law. Adopted community policies govern matters they validly address and may enter the agreement through incorporation. The application supplies facts, need and purpose where the contract makes them relevant. The bylaws define corporate powers and membership structure; they do not automatically create every operational restriction staff might consider desirable.

Board resolutions can administer the company and authorise enforcement within existing power. They should not become substitutes for community policy where the institutional constitution reserves resource-policy formation to the bottom-up process. Staff guidance can make procedures consistent, but unpublished practice is weak ground for terminating a member. Mission statements guide interpretation at the margins; they cannot contradict specific text.

Past practice may illuminate ambiguity. If AFRINIC knowingly accepted similar out-of-region use or customer models, that history could affect whether a duty was sufficiently clear, although consistent error would not permanently disable lawful correction. If AFRINIC repeatedly required approval for the same kind of change, that could support its reading if members had notice. The evidence should be portfolio-wide rather than selected from one dispute.

Where two sources conflict, the decision should identify the conflict and lawful resolution. A later policy should not be called an interpretation of an older contract merely to avoid temporal limits. A broad contract clause should not be stretched to displace a specific adopted policy. Clear hierarchy keeps institutional mission from becoming an unlimited reserve power.

It also gives the member a fair target. Cloud Innovation would have to answer the actual controlling term rather than alternate among policy, bylaws and stewardship rhetoric. AFRINIC would be able to defend enforcement as application of authority already granted. Courts and members could then review a legal chain instead of choosing between competing institutional stories.

What a defensible merits decision would contain

The opening section would authenticate the governing instruments: the July 2013 signed agreement, incorporated policies, allocation applications, amendment notices, renewal terms and the version effective in 2021. It would explain which clauses applied to which periods and why.

The second section would define each alleged breach separately. Eligibility, purpose, inaccurate information, non-cooperation, transfer, leasing, out-of-region use and failure to remedy would not be interchangeable. Each would have elements and a burden of proof.

The third would find facts by block and period. It would identify the original representation, quantity, actual use evidence, customer-control arrangement, communications and response. Technical uncertainty and sampling limits would be disclosed.

The fourth would address the member's case. It would answer the policy-silence argument, routing-method challenge, distinction between assignment and transfer, notice claims and evidence of customer need. Rejection would be reasoned, not assumed from AFRINIC's stewardship role.

The fifth would analyse proportionality and consistency. It would explain why termination, block-specific correction or another cure followed from the proved breach, how similar members were treated and what continuity measures protected downstream users.

The final section would state review rights and implementation dates. The evidential record would be preserved for a court or appeal body. Public reporting could summarise the rule and outcome without exposing customer information.

Such a decision might favour AFRINIC on some or all issues. Temporal discipline is not a prediction for Cloud Innovation. It is the standard that makes either outcome worthy of reliance.

The registry's power is strongest when its rule is least surprising

AFRINIC's mandate gave it both responsibility and leverage. It maintained the member relationship, registration, reverse DNS, routing-registry and certification services associated with the resources. In a scarce market, refusing to enforce clear commitments would harm members that followed the rules.

Power becomes fragile when the rule appears after the conduct. A member facing termination can turn an interpretation dispute into a governance crisis; courts must then untangle contract, company law and continuity under pressure. Other members cannot know whether their own arrangements are safe. Staff become policy makers by enforcement precedent.

The solution is not to weaken contracts. It is to write, preserve and apply them better. Every allocation should carry a versioned evidence schedule stating purpose, geography where relevant, customer model, reporting duties and incorporated policy. Every amendment should be delivered and acknowledged. Every review should use published measurement criteria. Every breach notice should be particularised. Every decision should be independently reviewable.

Cloud Innovation's case became explosive because institutional purpose, contract language, policy silence, technical measurement and commercial practice were allowed to collapse into one accusation. They needed to be separated. A clear purpose promise can be enforced. A clear future policy can govern future conduct. A proved false statement can justify sanction. None requires pretending that a withdrawn policy was already law.

The final lesson is modest and severe. Hindsight can reveal that an agreement was under-specified or a community rule arrived late. It cannot repair the past by changing what the parties promised. AFRINIC had to prove the obligation that existed, the fact that breached it and the fair consequence that followed. Anything less converted stewardship from the administration of rules into the assertion of institutional intention.

Sources and evidentiary boundaries

AFRINIC's 2017 Registration Service Agreement establishes the later published clauses concerning cooperation, purpose, adopted policy, breach, termination, renewal and appeal. AFRINIC's accompanying 2017 proposed-changes paper shows that this text changed from a January 2016 version. Neither document is treated as a substitute for the executed July 2013 agreement.

The archived out-of-region use proposal establishes its withdrawn status, its contemporaneous description of policy silence and its version history. A failed proposal is not an adopted rule; it is evidence of the ambiguity entities were attempting to resolve.

Cloud Innovation's 13 July 2020 response provides the asserted signing date, its technical objections and its interpretation of policy and contract. AFRINIC's litigation account provides the registry's competing purpose theory and chronology. Both are party statements tested against documents rather than accepted as neutral findings.

The Supreme Court of Mauritius July 2021 judgment establishes the preliminary representation issues on which the first injunction application was set aside. The later 2022 appellate judgment is used only to confirm that the appellate court expressly did not decide the remaining merits; it does not resolve the contractual dispute analysed here.