Summary

  • On 3 June 2010, at AFRINIC-12 in Kigali, the PDP Moderator Group chair recorded consensus on the IPv4 Soft Landing proposal subject to modifications. The surviving record identifies multiple requested changes but does not show a frozen meeting text, a clause-by-clause disposition, the editor of the post-meeting draft, or a redline connecting the discussion to the version placed into last call.
  • The qualification mattered because the disputed details could change access to the remaining unallocated IPv4 inventory: allocation ceilings, the denominator for out-of-region use, utilization conditions, IPv6 linkage, recovery treatment, end-user treatment, reserve discretion, and the duration of planning periods.
  • The strongest benign account is also supported by the record. Rough consensus can settle direction without settling every word; a later text was exposed for comment, objections were made, and the proposal returned to discussion. That is evidence of reversibility, not proof of fraud, capture, fabrication, bad faith, or invalidity.
  • The surviving gap is narrower and still important. A private registry can coordinate a queue for finite unallocated stock, but it must preserve the identity of the policy object as that object moves from meeting discussion to edited text. A public amendment ledger and immediate redline would have made the conditional act auditable without pretending that AFRINIC was a sovereign or a legislature.

A decision whose object was still moving

The AFRINIC-12 Public Policy Meeting convened at the Serena Hotel in Kigali, Rwanda, on 3 June 2010. The meeting minute names Vincent Ngundi, Paulos Nyirenda, Douglas Onyango and S. Moonesamy as discussion leaders. Douglas Onyango presented the IPv4 Soft Landing proposal. The official proposal archive records that an updated text had been posted on 12 May 2010 under the reference AFPUB-2010-v4-001. It then records the Kigali result in compact terms: the proposal reached consensus “with some modifications.” The meeting announcement similarly described Soft Landing as accepted with a few amendments required.

That compressed disposition is the institutional act examined here. The question is not whether people discussed IPv4 depletion, whether the room generally supported a softer path into exhaustion, or whether a later policy eventually took effect. Nor is it whether rough consensus must resemble a legislative roll-call vote. The question is what precise object the chair treated as having consensus when parts of the object were still expected to change.

The distinction is practical. Consensus may attach to a direction, a set of principles, a complete draft, or a base draft plus an amendment package. Each is a different object. If the public record says only “with modifications,” the people later holding the text—the author, chair and secretariat—must translate an oral and partly distributed understanding into operative clauses. Even honest editorial choices can alter meaning when the proposed rules contain numerical ceilings, eligibility definitions, geographic conditions and institutional responsibilities.

The record does not establish that this translation was manipulated. It establishes that the bridge is incomplete. There is no independently preserved, rendered and hashed copy of the exact 12 May text in the available record; no proof of the precise document projected or distributed in Kigali; no complete transcript; no public clause-by-clause resolution of the interventions; no named editor for the period between meeting and last call; and no comprehensive redline connecting the meeting object to the text made visible on 25 June. The missing bridge is a version-integrity problem, not evidence of a secret conspiracy.

That boundary matters because earlier Soft Landing episodes already have their own analytical lives. The initial draft, the May 2009 revision window, the first no-consensus result at AFRINIC-10, and the second no-consensus result at AFRINIC-11 are not alternative openings for this account. Prior BTW research has examined the adjacent May 2009 and AFRINIC-11 episodes. Their dates locate the Kigali event in a sequence, but their theses do not belong to this one. Here, the primary object is the conditional disposition on 3 June 2010 and the custody of changes required by it.

What the proposal was said to do

The meeting minute summarized the proposal’s stated objective as a strategy for allocating and maintaining AFRINIC’s final /8 from IANA. The author intended the mechanism to extend IPv4 availability while networks transitioned to IPv6. That was the proposal’s rationale as recorded and attributed to its author. It is not, by itself, proof that delaying administrative exhaustion would improve welfare, prove economic scarcity, or establish a continuing right of control over address space already in operational use.

This separation between proposal rationale and institutional conclusion is essential. A registry may face a real administrative problem when its stock of unallocated identifiers approaches a threshold. The need for an orderly queue can justify precise, thin coordination. It does not follow that every condition attached to that queue acquires the character of public regulation. Nor does the fact that the pool is finite transform the registry into the owner of networks, commerce, territory, or the identifiers’ operational value.

The 12 May update therefore entered Kigali as a policy instrument for a private service process. It contained or led into questions about who could receive remaining inventory, in what quantities, under what utilization threshold, with what geographic conditions, and with what relationship to IPv6. Those questions were not decorative. They determined the practical burden borne by LIRs, applicants and end users during the final phase of unallocated stock.

The archive’s later account of eventual implementation confirms that the Soft Landing approach became a phased inventory mechanism and that IANA allocated AFRINIC the final 102.0.0.0/8 on 3 February 2011. That is outcome context only. It cannot be read backwards to make a later implemented rule identical to the exact policy object assessed in the room on 3 June 2010. The very phrase “with some modifications” warns against that retrospective collapse.

The changes the room placed on the table

The minutes preserve enough of the discussion to show why textual custody mattered. They do not preserve enough to reconstruct a vote, an issue-by-issue consensus assessment, or a final amendment package.

McTim asked what would happen if an LIR used more than ten percent of its resources outside the AFRINIC region. He also objected that the proposal changed the philosophy of allocating according to need. Onyango responded that enforcement would be left to AFRINIC Ltd and that the design remained needs-based within new limits. These were the questioner’s concern and the author’s answer. They were not a community vote, and the answer does not prove that AFRINIC possessed public enforcement authority.

Alan Barrett then asked what the ten-percent allowance measured. Was it ten percent of AFRINIC’s total resources, or ten percent of each LIR allocation? Onyango answered that it was per LIR and said the policy would be updated to make that clear. A denominator can look like a small drafting detail, but here it changes the reach of the condition. Ten percent of a registry-wide pool and ten percent of each recipient’s block are materially different constraints. Once the author promised an update, the text after the meeting needed to reveal exactly how that promise was expressed.

Owen DeLong raised another implementation problem. If every IPv4 allocation automatically triggered an IPv6 allocation, an applicant could accumulate multiple unwanted IPv6 allocations. Onyango clarified that automatic IPv6 treatment was intended only for the first allocation. Again, the intervention yielded a verbal clarification whose dependable operational meaning required a clause. The record identifies the direction of the answer, but not a meeting-time redline adopted against a stable clause number.

Mouhamet Diop and John Walubengo asked when the policy would end and what would happen to IPv4 resources recovered as IPv6 use expanded. Onyango said the policy ended when the final /8 ran out and did not cover other IPv4 later obtained from IANA. Alain Aina asked whether recovered IPv4 would postpone exhaustion and how end users would be treated. The author said the end-user details had not been developed and that more input from the list was expected.

These exchanges exposed more than one unresolved edge. The end condition could depend on exhaustion of a particular inventory, on recovery, or on a later rule. Recovered space could return to the same queue, remain outside it, or demand a separate policy. End users could be treated like LIRs, given a distinct rule, or left unclear. The record shows questions and author responses, but it does not show a complete disposition declaring which answer became binding text on 3 June.

Aina also challenged the stated aim of extending IPv4’s lifetime. He reframed the purpose as preserving some IPv4 for possible future transition mechanisms. Mark Elkins said the ten-percent rule should operate per allocation. He reasoned that other regions would probably have exhausted their IPv4 pools by the time the rule applied and questioned the need to tie IPv6 expressly to Soft Landing. Onyango said the IPv6 linkage was meant to encourage uptake and acknowledged that IPv6 was not central to the proposal.

That debate matters because a purpose clause can guide interpretation of the operative text. “Extend IPv4’s lifetime” and “retain inventory for transition mechanisms” may lead administrators and applicants to understand borderline cases differently. Yet the minute does not say that the room adopted Aina’s reframing, retained Onyango’s original rationale, or treated the difference as immaterial. Nor does it show that Elkins’s reasoning about other regions was tested or accepted. Each remains the named participant’s position.

Barrett offered the most extensive recorded set of proposed changes. He supported applying the ten-percent rule per allocation. He recommended keeping the exhaustion phase in force even if space were recovered, until a new proposal dealt with recovery. He recommended removing IPv6 references, except perhaps a requirement that an applicant obtain IPv6 before further assignments. He questioned a proposed new responsibility for the Board to replenish a pool. And he suggested that the shift from a twelve-month to an eight-month planning period should apply in the current phase as well as in exhaustion.

Five suggestions from one speaker are not five adopted amendments. The minutes do not give a clause-by-clause ruling on Barrett’s package or on every other intervention. They do not identify whether the chair regarded each point as supportive, resolved, non-blocking, editorial, substantive, or outside the proposal’s core. They do not state whether any one concern was dispositive. They do not identify the revision completed before the consensus call, if a revision was completed at all.

This is why the phrase “the room accepted the policy” is too coarse. The record supports a more exact statement: the chair assessed consensus on the proposal subject to changes, after a discussion in which multiple changes were proposed or verbally clarified. It does not support a claim that every speaker supported the same final clauses. It also does not support the reverse claim that no consensus existed. The condition is part of the decision and must not be silently dropped from its description.

What the procedure allowed the chair to do

The published policy-development procedure then in force defined consensus as general agreement rather than a majority. It placed the determination in the hands of the Moderator Group co-chairs. If consensus existed at the meeting, the procedure called for a fifteen-day last-call period, during which comments agreed at the meeting were to be incorporated. After last call, the Moderator Group was to produce a report containing the proposal date, short summaries of the online discussion, face-to-face discussion and last call, and a recommendation to the Board.

That design makes post-meeting editing unsurprising. It supports the benign view that a meeting need not polish every line in real time. A chair can determine that the substantive direction has general agreement, allow an author to integrate clear comments, and expose the result to a further challenge period. Rough consensus is valuable precisely because it can coordinate technical judgment without pretending that every decision is reducible to fifty percent plus one.

But the procedure also makes the identity of “comments agreed at the meeting” consequential. Someone must decide which comments reached that status. Someone must choose the wording that incorporates them. Someone must decide whether an edit is merely clarifying or changes the bargain. Where the published record does not connect those choices to stable clauses, practical discretion moves from the visible discussion to the editorial chain.

That discretion is not automatically abusive. It is a dependency. The participants depend on the author, chair and staff to preserve the direction they heard. Applicants and absent principals depend on the last-call text to represent the conditional decision accurately. The Board, if later asked for ratification, depends on the report and text to state the object it is considering. Registry staff, if later implementing the rule, depend on the same chain to turn concepts into repeatable decisions.

The procedure proves the design of a private coordination process. It does not prove sovereignty. A Moderator Group’s power to assess general agreement comes from the process within which participants and the private registry coordinate. It is not legislative authority over everyone in a geographic region. A last call is a review mechanism, not public-law notice that converts a service rule into territorial jurisdiction.

The archive also records a timing fact that the short meeting disposition does not explain. Last call began on 25 June, twenty-two days after the Kigali meeting, although the procedure described a fifteen-day last-call period once consensus existed. The unknown is the reason for the interval. It could reflect editing, administration, deliberation, ordinary delay, or something else. The record does not permit a responsible conclusion. The interval matters only because it marks the period in which textual custody was especially important.

The text that appeared on 25 June

The last-call discussion reveals important features of the post-meeting proposal. A 25 June thread reproduces exhaustion-phase language setting a /24 minimum, a /23 maximum, and no more than four allocations or assignments. Graham Beneke asked for the resource-planning calculation behind the /23 ceiling. Scott Leibrand replied that the ceiling could ration medium and large ISPs after four /23s, encourage large customers to seek end-user address space directly from AFRINIC, and reduce aggregation. He did not say that those results were necessarily better or worse.

This exchange illustrates the stakes of version custody. A maximum block size and a four-request ceiling together shape more than a queue. They may change how an operator structures applications, how a large customer approaches the registry, and how address announcements aggregate. Even without any claim of sovereign power, the registry’s text can alter operational incentives because it controls access to the remaining unallocated inventory it administers.

A 27 June reply reproduces further provisions. It defines LIR, existing LIR, new LIR and end user. It requires ninety-percent utilization of all previous allocations or assignments before an applicant can receive more space. It limits each applicant to four separately requested blocks. It allows no more than ten percent of each exhaustion allocation or assignment to be used outside the region, solely to support connectivity back to the region.

The same reply points out wording and spelling changes and asks how the allocation size mentioned in the four-block provision relates to sizes in the current phase. It is one participant’s editorial and interpretive response, not a tally of consensus. Yet it shows that last call was doing real work. Definitions, cross-references and the relation between phases still required attention. The public was not merely blessing a document already beyond question.

Some meeting inputs can be compared with this later visible text. The ten-percent condition appears with a per-allocation or per-assignment denominator, consistent with the verbal clarification in Kigali. Definitions for different recipient types appear. The four-block, /24 and /23 constraints are visible. The meeting discussion had limited automatic IPv6 treatment to the first allocation. But comparison at this level cannot establish that every requested change was adopted, that every adopted change was captured, or that the 25 June text was identical to a single frozen text seen by the room.

Nor can it establish who wrote each provision. The author may have edited some clauses; chairs or staff may have assisted; existing language may simply have remained. The record supplied here does not allocate authorship. It also does not show whether the chair called consensus on the whole draft, on a principles package, or on a series of issues. A reader can identify areas of continuity, but not reconstruct a complete chain of custody.

The post-meeting text therefore supplies evidence without curing the entire gap. It shows an object that participants could inspect and challenge. It does not, without an amendment ledger or redline, prove precisely how the conditional consensus became that object. The difference is the heart of this inquiry.

The safeguard that did work

The archive records that last call ended on 14 July 2010 with objections and that the proposal returned to discussion. That outcome must be given full weight. It shows that the process remained reversible. The last-call text was not treated as untouchable merely because the meeting had produced a conditional consensus assessment. Participants could raise concerns, and unresolved objections could interrupt the path forward.

This is the strongest contrary case to a severe reading of the documentation gap. Rough consensus does not have to freeze every word in the room. The meeting may settle substantive direction; the author and chairs may incorporate sufficiently clear changes; last call then exposes the resulting language. On this account, the twenty-two-day bridge was an ordinary working interval, and the return to discussion is evidence that the review mechanism functioned when disagreement persisted.

That case is persuasive as far as it goes. The 3 June decision should not be called fraudulent, fabricated, void or captured simply because amendments followed. There is no evidence here of bad faith, corruption, unlawful conduct, deliberate manipulation, or a hidden bargain. There is also no complete record of the objections that caused the return, so it would be unsafe to nominate one issue as decisive.

Reversibility, however, answers a different question from traceability. Last call let participants object to the text they saw on 25 June. It did not necessarily let them verify, from the public record alone, that every accepted meeting modification had been incorporated faithfully or that no substantive addition had entered through editing. A reviewer who missed the Kigali session would need both the incoming text and a ledger of dispositions to perform that comparison reliably.

The safeguard thus worked in one important dimension: the policy object could be challenged and returned. The evidence remains incomplete in another: readers cannot trace every change from speaker, to chair assessment, to editor, to resulting clause. A mature account should hold both conclusions at once. The process was open enough to reverse course, and its surviving version-custody record was thinner than the decision required.

Who held practical power during the gap

The power chain began with an anticipated IANA final-/8 trigger and a proposal author preparing a pre-meeting text. It moved through discussion on the RPD list and at AFRINIC-12, including the interventions recorded in the minutes. The PDP-MG chair then assessed consensus subject to changes. From that moment until last call, practical control over the policy object sat with the people able to edit, approve and release the text. Once published on 25 June, the object returned to a wider review surface, where objections led back to discussion.

Only after such stages could later Board action and staff implementation affect applicants, LIRs, end users and operators.

The narrow legitimate control surface is easy to state. A private registry can maintain an accurate, auditable queue for its remaining unallocated inventory. It can record allocations, verify relevant conditions under its service arrangements, preserve uniqueness and publish dependable data. The contested surface begins where rationing ceilings, utilization gates, geographic-use conditions, IPv6 prerequisites, reserve discretion or enforcement language extend beyond what thin technical coordination requires.

Heng Lu’s first-person analysis, grounded in direct involvement across the RIR system, supplies the controlling authority framework. Participation contributes evidence, technical judgment, warning, objection and coordination. Attendance at a meeting, access to a microphone, presence on a mailing list and a rough-consensus assessment do not create a demos or confer authority over absent principals. The registry remains a private bookkeeper, ledger service and technical coordinator. It is not a sovereign, legislature, regulator, government, owner, police force, prosecutor, confiscator, punisher, court or adjudicator.

This authority boundary does not make registry work unimportant. On the contrary, recordkeeping can be operationally critical. Uniqueness, proof of control, accurate registration, reliable contact information, security assertions, transfer recording, dispute metadata, portability and continuity can justify centralized coordination. The test is what would break in the running Internet if a function were not coordinated. Commercial models, geographic worthiness, political approval and capital control do not become sovereign competences merely because they are expressed through an address-policy document.

Consensus earns technical weight while it serves the running system: uniqueness, interoperability, registry accuracy, security integrity and continuity. The order cannot be reversed so that an administrative record is treated as creating operational reality or overriding networks that are already functioning. The ledger describes and coordinates; it does not ascend into ownership.

On 3 June, the immediate inventory was still unallocated and the proposed mechanism concerned the registry’s handling of that pool. That fact supports a legitimate administrative role. It does not turn the consensus process into legislation. The policy could coordinate access to a private service and still produce meaningful economic and operational effects. Those effects are why its textual lineage deserved a stronger record, not why AFRINIC should be treated as a state.

Scarcity without mythology

Soft Landing was presented against administrative exhaustion of a free pool. Administrative exhaustion and economic scarcity are not identical. The former means that a particular registry’s stock of unallocated addresses has reached a defined stage or limit. The latter concerns the availability, demand, use and exchange value of resources across operational networks. A queue rule for distributing the remainder can respond to the first without resolving the second.

Confusing the two gives a temporary inventory mechanism too much conceptual reach. A /23 ceiling, a four-block limit or a ninety-percent utilization test may ration access to what the registry still holds. None of those rules creates title to all address space, permanent stewardship over resources after allocation, or territorial ownership of Internet operations. A geographic-use condition may be written into the queue, but the existence of the clause does not itself prove public jurisdiction.

This is also why the recorded “enforcement” answer requires care. Onyango said enforcement would be left to AFRINIC Ltd. The minutes prove that he said it; they do not establish a police, punitive or adjudicative power. In a private registry setting, enforcement must be interpreted, if valid at all, through the bounded service and recordkeeping relationship. The term cannot be allowed to smuggle sovereign authority into an administrative exchange.

The version gap could nevertheless have distributional consequences. Change the denominator in the ten-percent rule and a different amount of use is permitted. Change the maximum block size or number of requests and different applicants can obtain different quantities or face different transaction costs. Change the end-user definitions and a customer may apply through another route. Change the treatment of recovered space and the exhaustion phase may persist or end at a different moment. Change the IPv6 condition and access to IPv4 may become tied to a separate operational step.

Those consequences need not result from misconduct. They follow from text. A punctuation-level edit can be harmless; a revised denominator is not. The problem with an undocumented amendment chain is that outsiders cannot always distinguish the two. The cost appears as repeated interpretation, renewed list argument, planning uncertainty and suspicion that could have been avoided through more exact records.

The amendment ledger that should have existed

A defensible counterfactual does not require a parliament, regulator or command apparatus. It requires disciplined document custody.

At the opening of the meeting, the exact proposal shown to participants should have been frozen and hashed. Each clause should have carried a stable identifier. As discussion proceeded, a numbered issue ledger should have recorded the proposer, the clause at issue, the substance of the requested change, its rationale, whether it was editorial or substantive, and the chair’s disposition. Where wording was accepted, the ledger should have preserved the resulting language rather than only a paraphrase.

The chair could then have separated consensus on the base direction from consensus on substantive amendment packages. This would not demand a vote on every comma. It would identify which changes were within the assessed general agreement and which remained for last-call debate. If a matter such as end-user treatment was acknowledged as undeveloped, its status should have been marked open rather than silently entrusted to an editor.

Immediately after the session, the author, chair and staff should have released both a clean draft and a redline against the frozen incoming text. The release should have named the editor or editors, stated the publication time, and tied every changed clause to an issue-ledger entry. Dissent and unresolved concerns should have remained visible. Last call should have begun only once that exact object and its lineage were public.

At the close of last call, a further ledger should have classified objections as addressed, unresolved or dispositive, without converting them into anonymous noise. The fact that this proposal did return to discussion makes such a ledger especially useful: it would show why the safeguard activated while preserving the difference between a participant’s concern and the chair’s final process assessment.

Such records would have served several principals. Meeting participants could verify that their accepted clarification survived editing. Remote and absent members could determine what the room actually settled. Applicants could plan against the exact numerical and eligibility rules under review. A later Board could see the object presented for recommendation. Staff could implement clauses without reconstructing oral intent. Operators could separate stable requirements from proposals still subject to objection.

The ledger would also protect the author, chairs and secretariat. Where the record makes every transition visible, later critics have less room to infer manipulation from ordinary editing. Documentation is not merely a restraint on discretion; it is evidence that discretion was used within its mandate. The lack of a complete public bridge in this case does not prove misuse, but it leaves officials unable to answer some version-integrity questions with the record alone.

The design should remain reversible until the audit trail is complete. Reversibility is not institutional weakness. In a technical coordination process, it is protection against locking in a text whose consequences were not fully understood. The 14 July return to discussion demonstrated this principle in practice. The better design would pair that reversible path with contemporaneous traceability.

What can and cannot be concluded

The evidence supports a precise conclusion. AFRINIC-12 produced a real internal coordination act: the chair recorded consensus on the Soft Landing proposal, conditioned on modifications. The act was neither a final immutable text approval nor nothing at all. It moved the proposal into a phase where editorial custody had unusual importance.

The record identifies multiple matters that could have required modification or clarification. It later exposes parts of the text placed into last call. It shows objections and a return to discussion. It does not supply the complete frozen baseline, amendment-by-amendment disposition, exact editor, comprehensive redline, participation denominator, support count, opposition count, abstention total, complete transcript, chair’s detailed reasoning, or full list of dispositive objections.

Those absences do not justify saying that consensus was fabricated, that a policy was secretly imposed, or that the result was legally void. They also do not justify treating the last-call text as self-authenticating evidence of everything the room endorsed. The limited inference is that conditional consensus creates a custody dependency, and the surviving public record does not fully document that dependency.

The private nature of the institution sharpens rather than weakens the need for precision. AFRINIC could coordinate its unallocated stock and contractual service. It could not create public law, sovereign territory, criminal liability, confiscatory power or adjudicative jurisdiction. A strong amendment trail would keep the registry within its legitimate role: an accurate coordinator recording an intelligible, reviewable act.

The enduring lesson of Kigali is therefore not that consensus is inherently suspect. It is that the noun must have an object. When the object is still being modified, the institution must preserve who changed it, under what recorded instruction, and with what opportunity for objection. On 3 June 2010, the process captured direction. Between 3 and 25 June, the public record should also have captured the text’s custody.