Summary

  • AFPUB-2016-V4-001-DRAFT03, published on 22 July 2016 with the status Under Discussion, stated one revision from Draft 2: its proposed Phase 1 maximum fell from /15 to /18 after discussion at AFRINIC-24 and on the RPD mailing list.
  • The three-bit CIDR shift made the ceiling one eighth as large: 131,072 addresses became 16,384. That arithmetic is certain, but the record does not establish how many requests would have reached either maximum or whether the cut would have prolonged the pool.
  • Draft 3 retained a two-part exhaustion structure, a /11 Phase 2 trigger, a /22 Phase 2 maximum, no explicit minimum, an eight-month planning horizon, a 90-percent prior-use test, an IPv6 prerequisite, a regional-use condition, and /16, /14, and /13 reserves.
  • AFRINIC staff identified practical ambiguities involving triggers, critical-infrastructure definitions, reserve capacity, IXP treatment, valid requests, minimum-prefix handling, recurrence wording, and omitted rules. The page's notation Legal: None records the absence of a legal comment, not legal approval.
  • At AFRINIC-25 on 29 November 2016, Alain Aina and Omo Oaiya presented BIS. The authors acknowledged two competing proposals and the difficulty of consensus, sought a common problem statement, and requested a return to the list. The co-chair recorded No Consensus, a procedural result inside private coordination rather than a sanction, adjudication, or sovereign veto.

A smaller number, but no settled rule

The central event in Draft 3 can be expressed in one line, but its stakes cannot. On 22 July 2016, the authors of Soft Landing BIS published Version 3.0 and changed the proposed maximum allocation or assignment in the first exhaustion phase from a /15 to a /18. On 29 November, the proposal reached the AFRINIC-25 Public Policy Working Group session in Mauritius. It did not leave that meeting with consensus. Its presenters asked for it to go back to the mailing list, and the recorded disposition was No Consensus.

Those two dates frame the important conflict. The July revision showed that the size of the maximum was negotiable. The November outcome showed that negotiation had not produced a common mandate. Between them lay a serious administrative problem: a private registry responsible for an unallocated pool had to decide how it would process claims on finite inventory. Yet that practical need did not turn the registry into a legislature or turn a meeting procedure into regional government.

Draft 3's identity matters because loose language can make a proposal sound more settled than it was. The document was AFPUB-2016-V4-001-DRAFT03, formally described as IPv4 Soft Landing BIS Draft 3. Its page named Omo Oaiya, Joe Kimaili, and Alain P. Aina as authors and displayed the status Under Discussion. It presented itself as a text that would obsolete the existing IPv4 Soft Landing Policy if adopted. The conditional phrase is decisive: a proposal describing what it would replace is not evidence that replacement occurred. The archived Draft 3 record proves publication, wording, authorship, and status; it does not prove approval or operation.

The revision history is equally narrow. It says the Phase 1 maximum moved from /15 to /18 after discussions at the AFRINIC-24 public-policy meeting and subsequent discussion on the RPD list. The official proposal-history page confirms that version event. The closed record does not provide speaker counts, message totals, measured support, or the model that selected /18 rather than a nearby prefix. It therefore supports a specific proposition—one maximum was reduced—without supporting a broader story that the revision embodied the will of a region or solved the economics of exhaustion.

By the time Alain Aina and Omo Oaiya presented BIS at AFRINIC-25, the change was concrete but the legitimacy claimed for its effects was not. A maximum of 16,384 addresses would plainly constrain a maximum-sized request more than a ceiling of 131,072. What it would do to total demand, applicant behaviour, repeat requests, the duration of the pool, or access for later applicants depended on facts the proposal record did not quantify. The difference between a known rule and an inferred benefit is the difference between responsible description and institutional mythology.

What the CIDR change actually proves

CIDR notation makes the numerical effect exact. IPv4 addresses contain 32 bits. A /15 fixes the first 15 of those bits and leaves 17 bits for addresses within the block, producing (2^{17}), or 131,072, addresses. A /18 leaves 14 address bits, producing (2^{14}), or 16,384. Moving from /15 to /18 adds three fixed prefix bits. Each additional fixed bit halves the block, so three such steps divide it by (2^3): eight.

The proposed Phase 1 maximum was therefore reduced by 114,688 addresses per maximum-sized request. Put another way, eight /18 blocks fit into the address capacity of one /15. These are properties of the address blocks, not observations about allocation. They do not tell us that an applicant would have received a /15 under the prior proposal, that eight applicants would receive /18s under Draft 3, or that the same inventory would last eight times longer. A ceiling describes the largest permissible request under a proposed procedure; it is not a forecast of the size, timing, validity, or approval of actual requests.

That distinction is essential because scarcity policy invites seductive arithmetic. If the maximum is divided by eight, one might assume the pool serves eight times as many organizations or lasts eight times as long. But either conclusion would require demand data and behavioural assumptions. Applicants that needed more than a /18 might return later, change their deployment plans, seek address access elsewhere, or reduce immediate use. Applicants needing less than a /18 would not be affected directly by the lower ceiling. A queue with many small requests could behave differently from one dominated by maximum requests. The cap alone cannot reveal the mixture.

Nor does the arithmetic prove the authors' stated objectives. Draft 3 described an eight-month planning horizon and linked its approach to short-to-medium-term need and equitable distribution of the last pool. Those are design claims. To prove them, one would need at minimum the size and composition of the available non-reserved inventory, a demand distribution, the frequency of repeat applications, processing times, acceptance and rejection patterns, and scenarios under alternative ceilings. None of that appears in the record available for this event.

The numerical change nevertheless has real analytical value. It shows that the administrative scarcity boundary was chosen, not discovered. IPv4's 32-bit address space is a technical fact. The decision that a Phase 1 applicant should face a /18 ceiling is a procedural parameter. The former cannot be negotiated at a meeting; the latter can. The fact that the ceiling had previously been /15 and then became /18 after discussion demonstrates that the operating constraint was responsive to institutional choice.

This does not make rationing illegitimate. Any custodian of limited unallocated inventory needs some method of ordering and sizing requests. It does mean the method should be judged as a private service rule: Did the operator define it precisely? Did it disclose the assumptions? Could applicants predict how it would be applied? Was there a bounded correction path for errors? Could the operator show why this number served the stated purpose better than alternatives? The exact eightfold reduction is the beginning of that inquiry, not the answer to it.

The architecture that stayed around the cut

Draft 3 was not merely a number in isolation. It retained a two-part exhaustion design for the remaining pool. The proposed transition into Phase 2 would occur when no more than a /11 of the non-reserved portion of the Final /8 remained. Phase 2 would then carry a /22 maximum. In both phases, the text stated no explicit minimum. These elements determined when the tighter stage began and how small-request handling might interact with the registry's allocatable prefix sizes.

The proposal also used an eight-month allocation or assignment planning period. Instead of treating an applicant's long-range ambition as the unit of need, the design focused on a shorter forecast. For existing holders, it required at least 90 percent use of prior allocations or assignments before a request during exhaustion, while exempting a new Local Internet Registry or End User from that test on its first request. It further required an IPv4 requester to hold AFRINIC IPv6 resources, request them at the same time, or have IPv6 resources from an upstream provider.

Each mechanism addressed a recognizable queue problem. A short planning horizon can reduce the risk that one applicant reserves inventory for needs far in the future. A high prior-use threshold can discourage stockpiling by existing holders. An exception for a first request acknowledges that a new applicant has no prior block to utilize. An IPv6 prerequisite can tie access to the transition environment in which IPv4 exhaustion occurs. Yet recognizable intent does not establish operational success. Each requirement also creates questions about evidence, burden, timing, and unequal capacity to comply.

The regional-use condition introduced another layer. Draft 3 said AFRINIC resources were for the service region and that out-of-region use should be solely in support of connectivity back to the region. Read narrowly at the allocation desk, such wording could be an eligibility condition attached to requests from the operator's service constituency. Read expansively after allocation, it could be invoked as a continuing claim over how a network uses an operational asset. Those are not equivalent. A private registry may publish bounded terms for distributing its unallocated inventory.

It does not thereby acquire sovereign jurisdiction over customer geography, network strategy, routing, financing, or the continuing commercial life of addresses already allocated.

Three reserves further shaped the proposal. A /16—65,536 addresses—was designated for critical Internet infrastructures. A /14, containing 262,144 addresses, was reserved for new LIRs or End Users. A /13, containing 524,288 addresses, was reserved for unforeseen situations. Again, these figures are CIDR capacities, not proof that any reserved address was used, that the categories were correctly sized, or that a beneficiary received an allocation.

Reserves are economically meaningful because they withdraw option value from the general queue and preserve it for named future uses. That can be prudent. An infrastructure service or new entrant arriving later might otherwise find the pool depleted. But a reserve is only as defensible as its eligibility rules, demand assumptions, release conditions, and accounting. A broad category with a small reserve can create a contest among nominally eligible parties. A large reserve for an undefined future contingency can leave addresses idle while ordinary applicants face constraints.

A reserve without a transparent sunset or reallocation rule can outlive the premise used to justify it.

The retained architecture therefore combined several levers: a lower maximum, a phase trigger, a second-stage cap, a utilization test, a forecasting window, an IPv6 condition, a geography condition, and protected sub-pools. Looking only at the /18 change understates the operating surface. But claiming that Draft 3 changed all those levers would overstate the revision. The correct formulation is narrower: Draft 3's one stated change from its immediate Draft 2 baseline was the /15-to-/18 maximum, while the July text retained this broader rationing architecture.

That precision protects the record from two opposite errors. The first is to dismiss Draft 3 as a trivial edit because only one stated number changed. A ceiling that is eight times smaller can materially alter planning and application behaviour. The second is to attribute the entire architecture to the July revision. The archived Draft 2 page is relevant here only as the immediate /15 baseline; the present event is the Draft 3 adjustment and what happened to that adjusted proposal at AFRINIC-25.

Scarcity was real; the chosen constraint was administrative

It is possible to hold two thoughts at once. Unallocated IPv4 inventory was finite, and a private registry responsible for coordinating unique entries could not process requests as though the supply were infinite. At the same time, the particular ways an institution manages scarcity can create additional friction, privilege, and illiquidity. Technical scarcity does not sanctify every administrative response to it.

The distinction begins with the registry's core function. Accurate registration helps prevent conflicting claims and supports operational coordination. Uniqueness matters: the same globally routed address space cannot be assigned indiscriminately to multiple unrelated networks without consequences. A queue for unallocated inventory therefore needs validation, recordkeeping, and a rule for dealing with more demand than immediately available supply. None of this is controversial, and none requires sovereign status.

The analytical mistake comes when custody of a ledger is treated as ownership of the economy built around the entries. A registry's ability to decide whether a request satisfies published conditions for new inventory is not a general power to command the applicant's enterprise. A service region is not a state border. A policy mailing list is not a parliament. A co-chair's procedural determination is not a court judgment. The operator's useful coordination role remains useful precisely because it is limited.

Draft 3 sat at that boundary. Its strongest elements concerned the pool not yet allocated: maximum request size, a transition trigger, evidence of prior utilization, and reserved categories. These are matters a registry service must confront. Its more expansive implications emerged where allocation conditions threatened to follow addresses beyond the allocation transaction, particularly through broad regional-use language or undefined continuing expectations. Administrative necessity cannot be used as a bridge from queue management to permanent control.

The economics reinforce the need for limits. IPv4 addresses support deployed customers, network expansion, service continuity, and revenue planning. Operator analysis of address continuity explains why access to usable IPv4 space affects real operational decisions. But that context does not prove Draft 3 caused a delay, loss, or transaction. It shows why a change to the maximum deserved evidence proportionate to its potential effect.

Similarly, analysis of the cost of being late to a remaining pool illuminates the distributional stakes of caps and timing without supplying 2016 outcome data. A smaller maximum might allow more separate requests to be considered before inventory runs out. It might also require a growing network to reapply, bear repeated documentation costs, or seek alternatives earlier. Parties with experienced policy staff may navigate those cycles more easily than smaller operators even when the rule is facially equal.

This is why scarcity must not be discussed as a magic word that settles authority. The distinction between physical scarcity and policy-created illiquidity is foundational. There are only so many IPv4 addresses; that constraint is real. Yet eligibility filters, caps, recurrence rules, documentation demands, reserve categories, and geographic conditions determine how the remaining pool is experienced. Those choices can distribute inconvenience and opportunity. They require reasons and measurements, not reverence.

Draft 3's eightfold reduction made that point unusually visible. Nothing about the IPv4 bit length changed between Draft 2 and Draft 3. What changed was the proposed administrative maximum. A defensible institution would therefore show the bridge from observed demand to chosen cap, and from chosen cap to expected effects. Without that bridge, the number remains a contestable rationing judgment—perhaps sensible, perhaps not, but not a technical inevitability.

The staff questions exposed an unfinished instrument

The staff comments attached to Draft 3 are most useful when read as implementation diagnostics. They do not determine whether the proposal was wise, lawful, or representative. They identify places where the people expected to administer the text could not confidently translate it into consistent operations.

One question concerned the exhaustion trigger itself. If movement between phases determines when a stricter maximum applies, the trigger must be expressed so that inventory staff and applicants can identify the same moment from the same data. Ambiguity at that point can change which rule governs a request. A transition provision is not explanatory decoration; it is a switch with distributional effects.

Another cluster of comments concerned critical Internet infrastructure. Staff found multiple definitions rather than one stable category. That problem became concrete when considering top-level domains. The comments noted the possible inclusion of thousands of TLDs while the proposed /16 reserve could contain only 64 blocks of /22 size. The point was not that every TLD would necessarily request or qualify for a /22. It was that the relationship between the stated beneficiary class and the capacity set aside for it was unclear. If thousands may be eligible for 64 equal-sized units, administrators need ranking, sizing, exhaustion, and conflict rules.

Staff also identified inconsistency with IXP policy. A proposal cannot promise predictable administration if one clause points an exchange operator toward one treatment while an existing rule points toward another. The record does not establish which rule should have prevailed. It establishes that the interface was not resolved in the text before the meeting-stage discussion.

The meaning of a valid request raised a further issue. Queue rules often turn on when a request is deemed complete, eligible, or ready for evaluation. If that moment is undefined, administrative order can depend on discretionary judgments about missing material and cure periods. In a dwindling pool, placement in the queue can carry substantial option value. A definition that appears procedural can therefore affect who receives inventory.

Minimum-prefix handling was also uncertain because Draft 3 stated no explicit minimum in either phase. Networks do not receive arbitrary fractions of a prefix; routing practices, allocation boundaries, and applicable request categories shape what can practically be issued. If the remaining fragments are smaller than the ordinary minimum for a request, or if a justified need is below an expected size, the text should say how the operator responds. Silence transfers the decision from a published rule to implementation discretion.

The recurrence wording created another ambiguity. A smaller maximum can result in more repeat applications, making the interval and conditions for returning important. Does the eight-month horizon imply when a holder may seek more? How does the 90-percent utilization requirement interact with that timing? What happens when growth occurs faster than forecast? Staff's question did not answer these matters, but it correctly signaled that a cap without a clear recurrence rule is only part of an allocation procedure.

Finally, the comments pointed to missing rules for some critical infrastructure. Naming a protected class is not the same as building an administrable pathway. The proposal needed to explain eligibility, evidence, sizing, repetition, and treatment when the reserve was insufficient or when a service did not fit the enumerated cases. Otherwise, two similarly situated applicants might receive different answers from staff trying to fill gaps.

Taken together, these comments reveal a predictable risk in scarcity administration: undefined terms become hidden allocation power. Where the text does not determine the answer, staff must. That may be unavoidable at the margins, but a rule governing valuable and finite inventory should minimize those margins and make any residual discretion reviewable. It should not rely on institutional prestige to substitute for drafting precision.

The Draft 3 page separately recorded Legal: None. That phrase must be kept in its modest evidentiary place. It means no legal comment was recorded on the page. It does not mean counsel approved the text, that the text conferred public authority, that its regional-use condition was legally validated, or that the proposal could bind operators as law. Absence of a comment is not an affirmative opinion.

Nor were the staff comments a legal judgment in disguise. Their significance is operational. They show that, alongside the unproved benefits of a smaller maximum, administrators had unresolved questions about the proposal's switches, definitions, reserves, and request handling. A serious evaluation must include those questions because the quality of a scarcity rule is measured not only by its stated purpose but also by whether it can be applied consistently.

The strongest case for the /18 ceiling

The most persuasive defence of Draft 3 begins with the finite free pool, not with institutional grandeur. A registry holding unallocated inventory must distribute it somehow. If the maximum remains large, a few qualified requests could absorb substantial portions before later applicants arrive. Reducing the ceiling to /18 could preserve more decision points, allow more organizations to enter the queue, and reduce the effect of any single demand forecast on the remaining stock.

The surrounding mechanisms strengthen that case. An eight-month need horizon can narrow the claim to deployable near-term demand. Requiring 90-percent use of prior allocations or assignments can discourage an existing holder from accumulating a buffer while others wait. Exempting a first request avoids asking a new LIR or End User to demonstrate use of an allocation it does not have. The IPv6 prerequisite can ensure that an applicant seeking scarce IPv4 has at least engaged with the larger transition. Reserves can protect certain future applicants from a general rush on the pool.

On this reading, /18 was not a punishment for large networks. It was a queue-management device. Every ceiling will disappoint an applicant whose justified short-term demand exceeds it, but the registry must compare that burden with the possibility that larger issuances leave nothing for the next applicant. Fairness in a finite queue cannot mean satisfying every claimant without limit. It may instead mean publishing equal maximums, testing present utilization, and preserving room for entrants that have not yet appeared.

The AFRINIC-25 conduct of the authors also deserves its strongest interpretation. The official meeting minutes record two competing proposals and difficulty reaching consensus. The BIS authors expressed a preference for work on a common problem statement and asked that their proposal return to the list. The authors of the competing Overhaul proposal proposed a fresh joint effort and withdrew their proposal in hope of collaboration. This can be read as procedural restraint: rather than demand approval of their preferred instrument, participants recognized division and sought a shared foundation.

The minutes further record mention of a possible emergency PDP route if a projected first-quarter 2017 trigger made speed necessary. That was a procedural possibility discussed against a projection, not proof that the trigger occurred, that an emergency path was used, or that Draft 3 acquired force. Still, it illustrates the pressure the participants perceived: delay could matter when the inventory boundary appeared to be approaching.

This countercase should not be weakened into a caricature. A smaller ceiling can be a rational precaution. Reserves can preserve options. A return to discussion can be responsible. Private coordination is not illegitimate merely because it lacks sovereign force. In fact, voluntary and technically grounded coordination is often most credible when its participants understand the boundary of their role.

But the defence remains a hypothesis until its expected effects are modelled and its burdens examined. The record does not show projected pool-life extension under /18, the share of historical requests above that size, the likely number of repeat applications, the staff workload, or the distribution of benefits among existing holders, new entrants, critical services, and applicants with different administrative capacity. It does not show that /18 was better than /17, /19, a declining schedule, or a demand-responsive queue.

Most importantly, prudence in administering new allocations cannot create jurisdiction over old ones. The legitimacy of a cap on unallocated inventory ends where the justification ends. It may support transparent conditions for a request at the registry counter. It cannot support punishment, confiscation, or continuing control of an operator's already allocated assets and business. The strongest case for Draft 3 is therefore also a bounded case.

Who pays when the maximum shrinks

A ceiling distributes more than addresses. It distributes timing risk, paperwork, uncertainty, and the cost of adapting business plans. The direct numerical burden falls on a qualified applicant whose near-term need exceeds 16,384 addresses. Under the earlier proposed ceiling, such an applicant might have sought as many as 131,072 in one request. Under Draft 3, it would face a smaller maximum even if its documented eight-month need were larger.

That does not mean the applicant was entitled to a /15, nor does it prove the lower ceiling caused loss. It means the design shifted the balance between one applicant's immediate planning horizon and the option value preserved for later requests. The difference is economic even before an address changes hands because infrastructure plans are sensitive to available inputs, application schedules, and confidence about future access.

Repeat-request costs are one possible consequence. If an operator could justify more than a /18 over time, a smaller ceiling might require it to prepare and submit additional applications. Staff would evaluate those applications. The operator would demonstrate utilization and update forecasts. Each cycle could add delay and uncertainty. A well-resourced organization might absorb that work more easily than a smaller provider, complicating the assumption that an equal cap produces equal practical access.

There is also a planning tradeoff. A short horizon can prevent speculative reservation, but networks are not built instantaneously. Procurement, customer commitments, routing design, equipment installation, and geographic rollout can extend beyond eight months. If the policy recognizes only a narrow slice of a credible plan, the operator may have to choose between staged deployment and uncertain future access. That risk does not invalidate the horizon; it is a burden that should be measured.

The beneficiaries are also uncertain. A smaller cap may leave more inventory available for later applicants, including new entrants. Yet the reserve for new LIRs or End Users already carved out a /14. To evaluate the combined effect, one would need to know likely demand inside and outside that reserve, the eligibility criteria, release conditions, and how requests would be ordered. Otherwise, the general cap and the reserve may duplicate protections, leave gaps, or shift scarcity between categories.

The critical-infrastructure reserve has a similar distributional problem. A /16 holds 65,536 addresses, equivalent to 64 /22 blocks. If the eligible category could encompass thousands of top-level domains as staff observed, capacity alone cannot deliver broad coverage. The rule would need a sizing model and priority order. Without them, the reserve preserves theoretical option value while moving difficult choices into future administration.

The /13 for unforeseen situations carried the largest of the three reserved capacities: 524,288 addresses. Contingency reserves can protect continuity when a genuinely unexpected need arises. But “unforeseen” must not become an invitation to opaque discretion. A defensible reserve would specify who can authorize its use, what evidence is required, how decisions are recorded, when unused space returns to the general pool, and how applicants can challenge an administrative error without transforming the registry into a court.

The broader lesson is that fairness cannot be read directly from prefix length. A /18 ceiling may spread access, or it may produce more cycles for the same sophisticated claimants. It may protect newcomers, or reserves may already perform most of that work. It may slow depletion, or demand may rearrange itself. The record proves a substantial reduction in permitted request size. It does not prove the incidence of its costs or benefits.

That is why claims about equitable distribution and long-term regional interest remain claims. They are legitimate objectives to test. They are not self-authenticating evidence and cannot supply a mandate that the meeting record withheld. The proper response is neither to accept them because they sound public-spirited nor to reject them because the institution is private. It is to demand a transparent model that lets applicants and observers see who bears which burden under plausible demand scenarios.

What happened at AFRINIC-25

The 29 November 2016 meeting record is concise but sufficient to settle Draft 3's procedural fate at that stage. The PDWG session took place in Mauritius. Adewole Ajao chaired it, with Sami Salih participating remotely as co-chair. Alain Aina and Omo Oaiya presented Soft Landing BIS. Joe Kimaili was a named author of Draft 3, but the minutes do not name him as a presenter; precision about that distinction matters.

The discussion occurred in the presence of a competing approach. The minutes record the co-authors acknowledging that there were two competing proposals and that consensus was difficult. Rather than portraying their own clauses as the uncontested answer, they expressed a desire to work toward a common problem statement. The BIS authors requested that their proposal return to the mailing list.

The authors of the competing Overhaul proposal, according to the same minutes, proposed a fresh joint effort and withdrew their proposal in hope of collaboration. That fact supplies context for the request to continue discussion, but it does not license an account of the competing proposal's substance here. The important point is procedural: the participants recognized that agreement on the problem and instrument had not crystallized.

The co-chair record and meeting summary say No Consensus for Soft Landing BIS. The summary called for collaboration on a revised problem statement and proposal. The phrase is neither mysterious nor punitive. It meant that the private policy-development process did not record the agreement needed to move this proposal forward at that meeting stage. The proposal went back to discussion.

No consensus did not revoke an allocation, seize an asset, fine an operator, decide a legal right, or punish an author. It was not a judicial holding and not a sovereign veto. It also was not evidence that the proposal had failed on every substantive ground. A procedural body can decline to find consensus because objections remain, because participants differ about the problem, because drafting is incomplete, or because support is unclear. The minutes establish the result, not every thought in the room.

The meeting cannot bear the weight of regional representation sometimes placed on such events. Attendance at a policy session, participation on a mailing list, or contribution to floor discussion does not transform the gathering into a legislature for every person and network in Africa. The minutes are a record of a private coordination forum. They do not enumerate the views of all operators, prove proportional representation, or establish universal assent.

This limitation cuts both ways. Critics cannot treat No Consensus as a plebiscite rejecting every scarcity control. Proponents cannot treat discussion as an implicit mandate waiting to be formalized. The record shows a contested proposal, two approaches, an acknowledged difficulty, a request for continued work, and a no-consensus disposition. Anything stronger would be inference.

The authors' decision to return to the list should therefore be credited without being romanticized. It may show restraint and willingness to collaborate. It does not prove the eventual quality of a common proposal, the correctness of /18, or broad agreement on long-term regional benefit. Procedural modesty is valuable, but it is not outcome evidence.

The most important consequence of AFRINIC-25 for interpreting Draft 3 is negative and exact: the meeting did not supply consensus. That fact prevents the proposal's normative language from being restated as the choice of “the community” or “the region.” At this point in the record, Draft 3 remained what its page said it was—Under Discussion.