Summary

  • At 07:15:36 UTC on 13 May 2009, Douglas Onyango posted what he called the most recent Soft Landing text and invited comment before the following week’s face-to-face discussion. AFRINIC-10’s agenda placed the first Soft Landing presentation on 20 May and the policy discussion on 21 May. The common-review interval for that text was therefore seven calendar days to the presentation and eight to the discussion. The proposal itself was not new: a first version had been posted on 7 January and Onyango had published the complete comparison baseline on 13 January at 10:11:55 UTC.

    The policy question had circulated for months; the materially revised decision object had not.

  • Relative to 13 January, the May version replaced AFRINIC approval of an eight-month IPv6 adoption plan and a started-use condition with a requirement to show at least 90 percent use of previous exhaustion-phase allocations. It retained IPv6 co-allocation when an LIR lacked IPv6, made /23 the minimum and maximum allocation size, added a general limit of four additional allocations, left a separate new-LIR maximum of four blocks issued one at a time, removed the critical-infrastructure definition and /24 carve-out from the /16 reserve, and introduced an Africa-operations and regional-use condition.

    These were changes to access, evidence, pacing, beneficiaries and scope—not merely editorial corrections.

  • The strongest defence is also part of the record’s logic. The author was responding after months of discussion; a 90-percent gate could be more objective than registry ratification of an operator’s transition plan; a fixed /23 and repeated-allocation ceiling could preserve options for later applicants; co-allocation could remove an administrative obstacle to IPv6; and a geographic condition could seek to preserve a regional terminal pool for African operations. Revision was not itself suspect.

    The legitimacy problem was that immediate replies still found live ambiguity about whether the limit meant four total or one plus four, whether /23 should instead be a range, what pool the text covered, and what IPv6 conditions would actually do.

  • The 2008 AFRINIC policy-development text required at least 30 days of proposal discussion before a face-to-face meeting. Soft Landing easily had a proposal-level interval measured in months. No retrieved 2009 rule, however, establishes a version-specific minimum, a freeze, a redline duty or automatic deferral after a material amendment. The seven- and eight-day clocks therefore do not prove invalidity or a procedural breach. They establish a narrower point: the timing of a material version affects who can compare, model and challenge the actual mechanism under discussion.

    A defensible process would make that burden visible through a stable baseline, an exact redline, reasons by clause and a documented option to defer closure when common review is inadequate.

The act under examination

The controlled act is the 13 May publication, not the whole life of Soft Landing and not everything that happened at AFRINIC-10. The official proposal history supplies a limited chronology: Vincent Ngundi posted Douglas Onyango’s first draft on 7 January; Onyango posted an author revision on 13 January; and he posted another author revision on 13 May. The first draft matters here only to establish that chain. The complete 13 January message is the comparison baseline. The complete 13 May message is the act whose timing and content require analysis.

The account stops at the scheduled discussion clock; later versions, later decisions and implementation history do not answer the question posed by this revision.

That boundary matters because a version controversy can easily be distorted in two opposite directions. One distortion treats a proposal as unchanged merely because its title and overall objective persist. The other treats every amended sentence as a new proposal whose previous public life becomes irrelevant. Neither is adequate. The proposal-level subject—rationing a prospective final /8 to extend the useful availability of IPv4 during transition—had been visible since January. But the May text redistributed administrative judgment and applicant burdens in ways that an operator could not infer from the January title.

The honest chronology must preserve both facts at once: months of exposure to the subject and only a week of exposure to this material mechanism.

Onyango’s 13 January message said feedback had produced changes and identified unresolved questions, including how to define critical infrastructure and how usage should be measured. At 07:15:36 UTC on 13 May, he called the new text the most recent version and sought comments before the face-to-face discussion the next week. The official agenda then fixed the calendar reference points: a Soft Landing presentation on 20 May, seven calendar days after the post, and a Soft Landing policy discussion on 21 May, eight calendar days after it. These are calendar intervals, not estimates of reading time.

They say nothing about how many subscribers saw the message, how many compared the texts or how many affected operators were able to model the rules.

The official event report records 135 participants across AFRINIC-10 as a whole. That number cannot repair the missing denominator. It is not the attendance count for the Soft Landing session, a census of LIRs, a tally of reviewers, a count of authorized principals or evidence that all relevant operator types were represented. It tells us that 135 people participated in the event. Any stronger inference would convert event attendance into a mandate that the record does not supply.

A reconstructed diff, not an official redline

No official redline was retrieved. The exact comparison must therefore be reconstructed clause by clause from the two complete primary emails. That method can identify retained, removed, replaced and added mechanisms, but it must not be misdescribed as an AFRINIC-certified diff. The May email also omitted the January document-metadata block naming the organization, version as “Draft,” date of 5 January 2009, status and author. Its email header and sender still identified the post, but the body supplied no separate stable draft number or hash.

Nor does the record contain a certified copy proving that every sentence physically presented at the meeting was byte-for-byte identical to the email.

Several foundations remained substantially stable. The stated incentive was still to provide a flexible IPv4-to-IPv6 transition by extending IPv4’s useful life. The object remained the prospective final /8 AFRINIC expected under the global remaining-space arrangement. The two-phase architecture persisted: current allocation policy would continue until a request could not be fulfilled without using the final /8, or until fulfilling it would empty the pool outside that /8; then an exhaustion phase would begin. The planning and allocation horizon remained eight months rather than twelve.

The new-LIR subsection still contemplated a maximum of four minimum-sized blocks issued one at a time. A /16 reserve remained available for unforeseen future use, with an eventual return condition. The definitions of existing and new LIRs and the two affected-policy references also remained.

Those continuities explain why the May document was recognizably Soft Landing. They do not make the changes marginal. A policy mechanism is defined not only by its stated purpose but by its triggers, transaction sizes, eligibility tests, caps, exceptions, evidence and decision rights. On those dimensions, May changed much of what an applicant and administrator would have to do.

Context shifted from a conditional future to available space

The January text described the global remaining-space policy as still being developed and made Soft Landing conditional on that global policy’s ratification. By May, the text said the global policy had been ratified and framed Soft Landing as applying to address space that would be available through it. That was a status update with operational significance: the proposal no longer described only a contingent future arrangement. It was addressing the expected administration of a defined incoming terminal block.

The rhetoric of custody also narrowed. January said AFRINIC would hold the last block in custody and should use it acceptably. May stated that the document’s purpose was acceptable use by the AFRINIC community. Neither phrase proves ownership. A registry can record allocations, coordinate uniqueness and administer access to a still-unallocated pool under published rules and contracts. The ledger does not turn numerical identifiers into sovereign territory or AFRINIC property. The wording change is evidence of how the author framed the proposal, not a transfer of title or jurisdiction.

/23 became the fixed rationing unit

January described a /23 allocation and also linked allocations to the minimum size then in force. In the existing-LIR subsection, it allowed one minimum-sized allocation per application even if documented need was larger, followed by new applications when the existing criteria were met. It did not state a lifetime ceiling that applied to every LIR. It also contained an explicit sentence under which a later change in AFRINIC’s general minimum would automatically change the Soft Landing unit.

May made the rule starker: /23 was both the new minimum and the maximum allocation size, even when an applicant documented need for more. The explicit sentence automatically tracking a future change in the general minimum disappeared, although the new-LIR subsection continued to refer to the minimum size in effect. The existing-LIR subsection no longer contained the January one-per-application sentence or its closing reference to renewed eligibility under current criteria. Instead, the fixed /23 maximum operated generally, and repeat access depended on the new utilization gate.

The arithmetic is exact but limited. A /23 contains 512 numerical IPv4 addresses; a /24 contains 256. A /16 contains 65,536, and a /8 contains 16,777,216. These counts do not establish how many units are routable, actively used, economically valuable, transferable or owned. They also do not quantify the useful life that any rationing rule would add. Prefix arithmetic identifies the nominal scale of the administrative unit. It does not by itself demonstrate policy success.

A fixed /23 can be defended as a way to prevent one larger request from consuming too much of a terminal pool. It can preserve more future request opportunities and make each transaction comparable. The same rule can impose repeated-request costs on an operator whose documented near-term need exceeds 512 numerical addresses. It can shift risk from depletion into application frequency, processing delay and uncertainty about future eligibility. The record contains no demand model showing how often either effect would occur, so both remain mechanism-level possibilities rather than measured outcomes.

Two four-allocation formulations created an unresolved interaction

May added a general sentence saying that no LIR could receive more than four additional allocations after the exhaustion phase began. Elsewhere, the retained new-LIR subsection said a new LIR could receive a maximum of four blocks, issued one at a time. Those formulations are not obviously identical. “Four additional” can naturally suggest an initial block followed by four more. “A maximum of four blocks” suggests four in total. The text did not reconcile the general ceiling with the new-LIR total.

This was not an ambiguity discovered years later through abstract wordplay. Immediate replies asked about it. SM asked what the four additional allocations meant. Leo Vegoda asked whether a new LIR would receive one allocation plus four additional allocations or four in total. SM also asked how the one-plus-four figure had been derived. Their questions prove that readers encountered a live interpretive problem during the short interval. They do not prove which interpretation was correct, how broadly the concern was shared or what any later decision did with it.

The distinction mattered for capacity planning. Under one reading, a new entrant’s lifetime opportunity during the phase could be four /23 blocks. Under another, it might be an initial /23 plus four more. Because each /23 represents 512 numerical addresses, the two readings differ by one rationing unit, but the article cannot convert that arithmetic into a claim about customers served, economic value or real network capacity. The important fact is institutional: an applicant could not know from the text alone which ceiling would govern its sequence of requests.

A cap can preserve option value for operators arriving later. It can also deny further access to an operator that exhausts its sequence while still demonstrating need. Without applicant counts, request distributions, growth rates, processing times and a clear interaction rule, no source establishes winners, losers or pool-life extension. The May revision created the administrative choice; it did not supply the evidence needed to quantify its consequences.

The IPv6-plan gate was replaced by a 90-percent test

January used a qualitative and intrusive transition mechanism. An applicant seeking further IPv4 was to provide an eight-month IPv6 adoption plan, obtain AFRINIC’s ratification of that plan, receive an IPv6 allocation if it lacked one, and start using IPv6 under the ratified plan before obtaining additional IPv4. The text also addressed an applicant without an upstream path. This structure placed substantial judgment in the registry: staff would have to assess whether a plan was acceptable and whether use had begun.

May removed the plan, the promised eight-month implementation period, AFRINIC’s plan-ratification gate, the special no-upstream explanation and the started-use condition. It retained automatic IPv6 co-allocation where an LIR lacked IPv6. In place of the plan-and-use requirement, the text required an existing or new LIR seeking additional IPv4 to have used at least 90 percent of its previous exhaustion-phase allocation or allocations.

The replacement has a powerful institutional defence. Ninety percent looks more objective than asking a private registry to approve an operator’s business and engineering plan. It ties repeat access to demonstrated use of scarce prior allocations rather than a forecast about transition. Removing qualitative ratification can reduce the risk that staff preferences determine whose plan appears credible. Retaining co-allocation means the operator can obtain IPv6 resources without first satisfying a subjective planning review.

But a numerical threshold is not self-administering. The May text did not define the numerator, denominator, aggregation unit, observation date, accepted evidence, decision maker or review path. “Allocation or allocations” raises questions about whether use is measured block by block or across the applicant’s phase holdings. The text did not explain exception handling or how disagreement would be resolved. An undefined metric can move discretion rather than eliminate it: away from approving a narrative plan and into selecting evidence, timing measurement and deciding what counts as use.

Co-allocation also has a bounded effect. Making IPv6 resources available can remove an administrative barrier. It cannot prove that an operator will deploy IPv6, route traffic over it, have compatible systems, secure business feasibility or complete a transition. SM’s immediate reply distinguished allocating IPv6 from adopting it and argued that AFRINIC could not tell operators how to run their networks. Graham Beneke, from a different direction, wanted a concrete deployment condition. Those conflicting comments reveal the trade-off between avoiding intrusive plan control and demanding evidence of meaningful transition.

They are individual positions, not a representative verdict.

The reserve lost its named critical-infrastructure channel

Both versions preserved a /16 for unforeseen future use and contemplated its eventual return. A /16 contains 65,536 numerical IPv4 addresses, but that count alone says nothing about its usable value or eventual effect. January additionally defined critical-infrastructure providers and allowed /24 assignments to them from the reserve. May removed both the definition and the /24 critical-infrastructure carve-out, leaving the /16 framed only around unforeseen future use and return.

Removal can be defended as administrative simplification. The January message itself identified the definition of critical infrastructure as an unresolved question. Eliminating a contested beneficiary category avoids having the registry decide whether a particular service qualifies for special treatment. It also changes the access surface. Applicants who might have relied on an express /24 channel no longer had that named route in the May text. The record does not show who would have qualified, how many requests might have arisen or whether the reserve would last longer. It shows only that the beneficiary architecture changed.

A new regional-use condition widened the eligibility inquiry

May added language saying AFRINIC resources were for the AFRINIC geographic region, that none could be used outside it, that requesting LIRs must have operations in Africa, and that allocations must support African operations. The January baseline did not contain this package of conditions.

The strongest charitable reading is preservation. If the terminal pool was expected through a regional registry, a geographic condition could seek to keep the remaining opportunity available to networks serving African operations. It could deter applications with no meaningful connection to the service region. Yet the wording left the unit of assessment unclear. Would the test apply to the applicant’s legal presence, infrastructure, customers, services, the specific allocation, or every route using the allocation? How would it treat a multi-region operator or a service whose infrastructure and users cross borders?

What evidence would establish compliance, and what consequence would follow for a new request?

These are questions of narrow allocation administration, not territorial sovereignty. AFRINIC is a private bookkeeper and coordinator. It is not a government, legislature, regulator, police force, prosecutor, court, punisher, confiscator or owner of number resources. A clause governing eligibility for a new allocation from a still-unallocated pool does not grant jurisdiction over companies, networks, territories or issued holdings. The legitimate control surface is the registry’s own allocation and contractual process. Describing it more grandly would inflate a service-region rule into powers that no source establishes.

Planning language changed while the eight-month horizon stayed

Both versions reduced the planning or allocation horizon from twelve months to eight. January justified the reduction as a way to minimize waste while others suffered deficiency. May reframed it in terms of short-to-medium-term need, fairness and equitable distribution. The horizon therefore stayed constant while its rationale changed.

This difference illustrates why language and mechanism must be separated. “Fairness” does not measure distribution, and “waste” does not define utilization. The sources provide no empirical estimate of how a shorter horizon, fixed /23, four-allocation ceiling or 90-percent gate would interact. There is no proven extension of pool life, no identified applicant distribution and no measured administrative cost. The defensible statement is that the mechanism was designed to pace requests and preserve future options; its realized effects remain unknown.

Immediate replies exposed the work still required

Four contemporaneous replies on 13 and 14 May supply unusually useful evidence because they connect the short clock to specific clauses. They should be used carefully. SM asked what “four additional allocations” meant and questioned why IPv6 belonged in the proposal. Leo Vegoda asked whether the policy applied only to the final /8 or to other AFRINIC stock, raised the four-total versus one-plus-four problem for new LIRs, and requested comparison with membership figures.

Graham Beneke argued that a fixed /23 was too restrictive, proposed a range from /24 to /20, pointed to both very small operators and those with needs around /21, advocated a concrete IPv6 deployment condition, and asked for LIR-count and growth data. SM then emphasized scarcity and the risk of slower growth, questioned the derivation of one plus four, distinguished IPv6 allocation from actual adoption, and rejected the idea that the registry could direct operators’ network management.

These messages do not constitute a vote. They are not a statistically representative sample, a census of affected operators or evidence of consensus. The record does not reveal the number of RPD subscribers, actual readers, people who compared versions, eligible participants, represented organizations or authorized principals. It would be equally wrong to treat four replies as proof that the revision failed and to dismiss them because they were only four.

Their proper evidentiary role is narrower: within roughly a day, identified readers found unresolved questions about cap interaction, scope, transaction size, empirical support and the relationship between IPv6 allocation and deployment.

That evidence bears directly on version timing. A redline would have let readers see quickly that the plan-ratification system had disappeared, a 90-percent gate had arrived, the critical-infrastructure exception had vanished and the geography clause was new. Reasons by clause would have explained which comments or facts motivated each move. A stable identifier would have fixed the object being reviewed. None of these devices guarantees agreement. They reduce the cost of discovering what requires disagreement.

The 30-day rule and the missing version rule

The archived 2008 AFRINIC policy-development process required at least 30 days of proposal discussion before presentation at the face-to-face meeting. The first post on 7 January and the May agenda plainly place the Soft Landing proposal beyond that proposal-level minimum. It is therefore inaccurate to say the subject was introduced only seven or eight days before AFRINIC-10.

It would also be inaccurate to infer a rule that the record does not contain. No retrieved 2009 source states a minimum interval for a materially amended version. None establishes a freeze date, a mandatory redline, an automatic deferral rule or a duty to restart the full 30-day proposal clock after changes. The record does not prove that the May timing violated the operative process, caused any outcome or was chosen strategically. A legitimacy critique must not be smuggled into a false allegation of procedural invalidity.

The better distinction is between proposal notice and decision-object notice. Proposal notice tells participants that a policy subject exists. Decision-object notice gives them a stable set of operative conditions to compare, model and authorize. A months-old title may satisfy the first while a last-week rewrite weakens the second. Whether eight days is enough cannot be answered by ritual. It depends on materiality, interaction among clauses, affected operator types, availability of a redline, access to data and what the meeting is expected to do. Here, the immediate questions show that the May text required real interpretive work.

Participation collected evidence; it did not manufacture authorization

AFRINIC’s own PDP overview described an open process using a public list and a face-to-face meeting, coordinated by the PDP-MG. Public archives and open invitations are meaningful institutional goods. They allow objections to be stated, expose drafts to expertise and make later audit possible. An open process is preferable to a closed allocation decision in those respects.

Openness does not identify the principals for whom participants speak. A mailing-list contributor may speak for herself, an employer, a member that has expressly authorized her, or no one beyond the argument she submits. A meeting attendee does not automatically become an elector for absent operators or a representative of Africa. The phrase “AFRINIC community” records institutional vocabulary; it does not prove a polity, legislature or authorization chain. The event-wide count of 135 cannot supply the missing link between attendance and mandate.

Heng Lu’s doctrine is controlling on this boundary but not evidence of the 13 May event. It treats registry records as coordination instruments: they describe allocation and support uniqueness, but do not create sovereign title, ownership or control of routing reality. It also rejects the conversion of service-region participation into public authority merely through attendance. No direct event-specific material from Heng Lu, NRS, LARUS or BTW establishes what occurred in this revision. Event facts come from AFRINIC’s official materials and contemporaneous messages; doctrine supplies the rule for interpreting the limits of those facts.

The distinction protects both the registry and affected operators. It lets AFRINIC administer a scarce unallocated pool without pretending to regulate society. It also prevents critics from misdescribing every allocation condition as state coercion. A private coordinator can set disclosed eligibility conditions within its service and contractual surface. Its remedies must remain within that surface. It has no legislative, police, prosecutorial, judicial, punitive or confiscatory power, and it does not own the numerical resources it records.

The strongest institutional defence

A fair assessment begins with the best case for May, not the easiest accusation against it. The proposal had circulated for more than four months. The author’s decision to revise can be understood as proof that comments were capable of changing the document. A process in which text never changes after consultation would be performative, not responsive. The May post expressly invited more comment rather than concealing the revision.

Several changes also had administrable logic. The 90-percent gate could replace an opaque judgment about the credibility of an IPv6 plan with a repeatable utilization test, at least if measurement were later defined. Fixed /23 transactions and a cap could ration the terminal pool and protect opportunities for later entrants. IPv6 co-allocation could make transition resources immediately available without requiring the registry to approve business strategy. Removing the critical-infrastructure carve-out could avoid a difficult category judgment while preserving the /16 for genuinely unforeseen uses.

The geographic clause could seek to align the regional terminal pool with operations in the region. The public list and meeting exposed these choices to challenge rather than keeping them internal.

The timing defence is straightforward too. Authors need freedom to improve a draft as evidence arrives. A rigid freeze can preserve a known defect, delay an urgent correction or force participants to discuss language everyone already expects to replace. Scarcity raises the cost of delay as well as the cost of haste. Keeping January as the only meeting text would have retained the qualitative plan-ratification system and the unresolved critical-infrastructure definition even though May offered possible improvements.

Nothing in the retrieved record justifies attributing bad faith, manipulation or hidden intent to Onyango, Ngundi, AFRINIC staff or the commenters. Revision can be a good-faith response to public engagement. Any serious recommendation must preserve that possibility.

Why the defence does not settle legitimacy

Responsiveness and review are complements, not substitutes. A draft can improve because it changed and still require more common time because the changes are material. In May, the decision object changed across almost every operational dimension: fixed transaction size, repeat-allocation ceiling, utilization evidence, IPv6 conditions, reserve beneficiaries and geographic scope. These provisions interacted. The meaning of a four-allocation ceiling affected how the /23 maximum would accumulate. The 90-percent gate affected when each repeat request could occur. The geographic condition affected who could enter that sequence.

Removing the protected /24 path changed how the reserve related to ordinary allocations.

The immediate replies show that the interpretive burden was not hypothetical. Readers disagreed about four total versus one plus four and fixed /23 versus a range. They asked what pool the rule covered, what evidence justified the cap and whether IPv6 allocation would produce deployment. The record supplies no published review denominator and no authorization chain from affected operators. Open access meant people could comment; it did not prove that materially affected parties received equal practical opportunity to understand and test the text.

The correct remedy is procedural design, not retroactive condemnation. Publish a stable baseline. Supply an exact redline. Explain the reason and intended effect of each mechanism change. Identify unresolved interactions. Give participants enough time to model applicant scenarios. If that cannot occur before the scheduled meeting, preserve the meeting as an evidence-gathering opportunity while deferring closure on the new text. Such a framework does not make amendments improper. It makes responsiveness auditable.

What can and cannot be concluded

The record proves a precise act and a precise clock. Onyango published the most recent version on 13 May. The agenda placed a presentation seven calendar days later and policy discussion eight days later. The comparison with 13 January proves material changes, and the immediate replies prove that identifiable readers encountered unresolved ambiguity and trade-offs during that interval. The process text proves a proposal-level 30-day requirement, which the proposal satisfied through its months-long history.

The record does not prove a version-freeze violation, procedural invalidity, strategic timing, representative approval or a causal outcome. It does not reveal how many people reviewed the May text, how many organizations they represented, who authorized them or how many affected operators remained absent. It does not resolve the two four-allocation formulations, define 90-percent use, specify the regional-use test or identify an exception and review process. It does not prove pool-life extension, applicant winners, routing effects, IPv6 deployment, business costs or enforcement outcomes.

The resulting judgment is deliberately narrower than either institutional celebration or procedural indictment. The May revision may have been an honest and technically motivated improvement. Its materiality made the last-week publication consequential because it compressed the shared opportunity to examine the actual rules. Version timing is therefore part of consensus legitimacy even when no written freeze has been breached. Legitimacy here means an inspectable basis for a private coordinator’s allocation choice—not sovereignty, legislation or ownership, and not a mandate manufactured by list access or meeting attendance.