Summary

  • The ARIN and APNIC feedback published on 9 October 2020 addressed a source-compliance sentence found in Draft 2 and Draft 3. Draft 4, dated four days earlier, no longer contained that sentence. The feedback proved a defect in the wording it tested, but did not by itself establish whether Draft 4 was compatible or incompatible.
  • A serious compatibility ground therefore had to do more than show that questions remained. It needed to connect the exact challenged text to the exact counterpart policy, effective date, transfer direction, resource type and dated interpretation, then show how the resulting gap could have affected the 7 October consensus call.
  • The Appeal Committee’s 23 December 2021 result settled a procedural issue in the final-review chain and annulled the 7 October decision. It did not decide Draft 4’s compatibility, endorse every allegation in the appeal or make the requested finding of bias or unfairness.
  • The right institutional response is a version-controlled compatibility receipt. That record can improve the work of a private technical bookkeeper and coordinator, but neither consensus nor internal review gives AFRINIC public authority.

The revealing fact in this dispute is not simply that feedback arrived late. It is that the feedback arrived carrying the grammar of an earlier draft.

On 7 October 2020, AFRINIC’s co-chairs announced that consensus on AFPUB-2019-V4-003-DRAFT04 had been maintained at the end of last call and that the proposal would be recommended for ratification. Two days later, staff published partial feedback from ARIN and APNIC. Both responses identified a difficulty in a sentence requiring the source of a transfer to comply with the receiving registry’s policies. Yet the official Draft 4 text, dated 5 October, had removed that sentence.

The public record therefore held two propositions at once: counterpart registries had identified a genuine problem in wording sent to them, and the latest draft no longer used those exact words.

That is not a technicality. It is the centre of the evidentiary problem. If the older sentence remained relevant in operation despite its deletion, the record needed to show how. If the amendment cured that problem, the record needed to identify what had been cured and whether any new directional gap appeared. If nobody had yet tested the latest wording, the honest entry was “pending”, not “compatible” and not “incompatible”. A compatibility conclusion without those distinctions would ask a label to do the work of a clause-by-clause analysis.

The appeal dated 20 October challenged the 7 October determination. Alain Patrick AINA was named as appellant, supported by Marcus K. G. Adomey, Vivien Assangbe Wotto and Mirriam Lauren. The filing alleged failures in administering the Policy Development Process, substantive amendment during last call, unresolved objections, incomplete compatibility work, and bias or unfairness in the handling of competing proposals. Those descriptions matter because they define what the appellant asked the internal process to review. They must nevertheless remain attributed allegations.

An appeal is not self-proving, and its requested remedy is not a finding.

For the compatibility ground in particular, the strongest question was narrower than the filing’s entire catalogue: what must an appellant prove when the objection cited against a transfer proposal belongs to one version, while the decision under review concerns another? The answer starts with the immutable text, but it does not end there. Inter-registry compatibility is directional. It depends on who sends, who receives, what resource is transferred, which policy applies at each end, and how each registry understands the practical obligations imposed on its account holder. It also depends on time.

A reply dated against one text cannot silently migrate to a later text merely because the proposal number looks familiar.

This is why the episode deserves treatment as a story about evidence architecture rather than a referendum on a policy outcome. The record offers enough to identify the gap, but not enough to fill it retrospectively. It shows that the proposal moved; that a known source-compliance concern was being checked; that staff had contacted four counterpart registries; that partial responses were published; and that further updates and a check against the latest version were promised. The sealed material does not contain a conclusive, dated assessment from all counterparts testing Draft 4.

That absence is evidence of an incomplete record, not proof of either substantive answer.

The objection arrived with a version number

The sequence matters because every step changes what a later assertion can support. On 17 September, at AFRINIC-32, the co-chairs declared conditional rough consensus on the Resource Transfer Policy, subject to specified amendments. Their 21 September summary described it as the least contested of three competing proposals, acknowledged a concern involving reciprocal treatment of legacy-resource transfers with ARIN, and announced last call. On 23 September, after the authors sent an updated version, a co-chair said the co-chairs would examine it for reciprocity and asked staff to assist.

That exchange establishes a live compatibility question. It does not establish the answer. It shows that the chairs recognised a need for checking, and that outside implementation knowledge from another registry could inform the inquiry. In a system of distinct registries with distinct rulebooks, this is sensible administration: the counterpart is often best placed to say how its own workflow would treat a proposed obligation. But the evidentiary value of that statement depends on preserving what, exactly, the counterpart reviewed.

Draft 4 was dated 5 October. Its official page recorded changes to sections 5.7.3.1, 5.7.3.2, 5.7.4.1 and 5.7.5. Section 5.7.3.1 required a source to be the current and rightful holder of resources registered with any regional internet registry and required those resources not to be in dispute. The source-compliance sentence quoted in the later ARIN and APNIC feedback—requiring the source to comply with the receiving registry’s policies—was no longer there.

The 7 October end-of-last-call message then said consensus had been maintained. The 9 October staff message reported only partial feedback. According to that message, ARIN considered the earlier source-compliance sentence incompatible because it might require a source in the ARIN region to comply with AFRINIC policy. APNIC identified the same issue in Draft 3. Staff said all four other regional registries had been contacted, that further information would be shared and that consistency against the latest proposal version would be checked.

The dates prevent three common shortcuts. First, the ARIN and APNIC observations cannot fairly be described as findings that Draft 4 was incompatible. Their identified defect attached to earlier wording. Secondly, deletion of the sentence cannot fairly be described as proof that Draft 4 was compatible. A clause may be cured, while a different interaction elsewhere remains uncertain. Thirdly, a missing response from RIPE NCC or LACNIC cannot fairly be converted into assent. Silence may mean that a reply was pending, that the question was still being framed, or simply that no conclusive answer is present in the sealed record.

The useful unit of evidence is therefore not “ARIN feedback” in the abstract. It is something closer to: ARIN; response date; proposal Draft 2 or Draft 3; section and quoted sentence; outbound or inbound flow; IPv4 or ASN; counterpart policy and effective date; interpretation; expected operational consequence; status after Draft 4. The same discipline applies to APNIC and to each other counterpart. Without that envelope, even an authentic reply can be misapplied.

Version control is more than the proposal number in a subject line. It requires a preserved text or checksum, a clause identifier, a record of the change and a link between the reply and what the respondent actually saw. A statement such as “the proposal was checked with other RIRs” collapses too much. It does not tell a reviewer whether the check concerned an earlier draft, one transfer direction, only IPv4, a legacy-resource edge case, or the entire operating path. Nor does it say whether a later edit resolved the problem or merely changed it.

The distinction resembles a chain of custody for reasoning. Nobody needs to doubt the good faith of staff, chairs, authors or counterpart registries to insist upon it. The point is to prevent evidence from losing its boundaries as it passes through a fast-moving process. A dated reply is reliable for the question it answered. It becomes unreliable only when the record asks it to answer a materially different question without saying so.

That boundary also protects the appellant. If a filing cites a genuine adverse answer tied to Draft 3, the internal reviewer can acknowledge the answer without treating it as automatically dispositive of Draft 4. The appellant can then identify whether the challenged obligation survived in another form, whether removal created uncertainty about which side’s rules controlled, or whether the failure to obtain a fresh response mattered at the moment consensus was declared. The task becomes proof, not insinuation.

Compatibility is a route through two rulebooks

It is tempting to treat inter-registry compatibility as a single yes-or-no property. The practical inquiry is more demanding. A transfer has a source and a recipient. Each may sit within a different registry’s service region and relationship framework. The process may concern IPv4 space, autonomous system numbers or a legacy-status question. One end may test the source’s authority and the status of the resource; the other may require the recipient to demonstrate need. A policy can look coherent when read within one registry and still create an impossible or ambiguous sequence when paired with the other registry’s current rules.

Draft 4 illustrates why direction matters. It required recipients of transfers into AFRINIC to demonstrate need under AFRINIC policy. It also said that transfers from AFRINIC to another registry would follow the relevant policies. Those propositions do not supply a universal answer. A transfer into AFRINIC and a transfer out of AFRINIC present different allocation of duties. The relevant counterpart rule might also distinguish resource types or statuses. A broad label of compatibility can conceal a confirmed path in one direction and a pending question in the other.

The appeal burden should therefore be understood as a set of linked propositions. The appellant must first identify the challenged clause in the version covered by the decision. Next comes the counterpart rule said to conflict with it, including that rule’s version and effective date. The filing should identify the direction and resource type, rather than relying on the generic word “transfer”. It should preserve the counterpart registry’s dated interpretation or, if no interpretation exists, state that the claimed mismatch is the appellant’s reading.

It should then explain the operational consequence: which requirement cannot be met, by whom, at what point and with what effect on completion or registry recording.

Finally comes materiality. The historical policy manual gave an appellant a means to challenge a chair action after prior discussion, with the required support and within the applicable time. That mechanism did not convert every unanswered query into a decisive defect. For compatibility, an appellant needed to show why the particular gap could have altered the consensus assessment, made the version presented for decision insufficiently reviewed, or left a material objection without an evidential disposition. The bridge from uncertainty to the challenged decision is not optional.

This burden is demanding for a reason. An internal appeal body should not have to become the author of the policy or improvise a cross-registry implementation design. Its job is to examine a bounded private record and apply the remedy available within that process. If the filing provides only a general anxiety about reciprocity, the reviewer risks substituting its own technical view. If it provides a versioned matrix and a materiality account, the reviewer can ask whether the chairs had a supportable basis for their stated decision without rewriting the draft.

The same structure disciplines the response. Chairs can answer each row: confirmed; conditional; pending; superseded by amendment; disputed; or immaterial, with reasons. They can show when staff asked a question, what text was supplied, which answer arrived before the decision and which arrived afterwards. They can explain why an unresolved combination did or did not matter. This is stronger than claiming that consultation occurred, because it reveals both the reach and the limit of consultation.

The record from October 2020 leaves the narrow compatibility conclusion open. ARIN and APNIC supplied adverse interpretations of an earlier sentence. Draft 4 removed it. Staff described the replies as partial and promised more information plus a latest-version consistency check. From those facts, one may infer that the evidentiary hand-off between versions was incomplete. One may not infer the ultimate compatibility of Draft 4. This distinction is the difference between auditing a decision and choosing a preferred policy answer.

A missing reply is a status, not a conclusion

Policy processes often reward compressed language. Chairs need to summarise long discussions, participants need to decide whether an objection remains live, and staff need to turn correspondence with counterpart organisations into a usable record. Compression becomes dangerous when it removes state. “Consulted” may hide that a request was unanswered. “Compatible” may hide that only one direction was tested. “Issue resolved” may hide that the criticised sentence disappeared but its operational purpose migrated elsewhere.

A compatibility ledger should preserve uncertainty as data. Each combination of proposal clause, counterpart registry, direction and resource type should carry a visible state. “Confirmed” means that the current wording was checked against the current counterpart rule and received a dated, intelligible answer. “Conditional” records assumptions or operational prerequisites. “Pending” means that the question was sent but not conclusively answered before the cut-off. “Superseded” means that the response belongs to text that changed, with a separate entry needed for the replacement.

“Disputed” preserves conflicting interpretations rather than hiding them in a narrative summary.

This approach does not require unanimity from every registry on every drafting choice. Nor does it outsource AFRINIC’s private policy process to counterpart organisations. The chairs and participants can still judge the proposal before them. What it prevents is a factual ambiguity about the evidence used. A counterpart’s explanation of its own workflow is valuable implementation evidence; it is not a veto, and it cannot make AFRINIC more than a registry coordinator. Conversely, a drafting choice within AFRINIC cannot dictate how another registry interprets its own policy.

Treating “pending” honestly can also improve pace. Not every unanswered point is material. If a row concerns an impossible or irrelevant direction, the chairs can explain that. If an issue affects a common transfer path or makes requirements mutually unsatisfiable, more review time may be warranted. The ledger separates the decision about materiality from the fact of whether a response exists. That is more efficient than allowing participants to argue over a global label without a shared map.

In this case, the timing cut-off should have been explicit. The final consensus announcement was dated 7 October; the partial feedback was published on 9 October. A decision record could have said what counterpart evidence was available by 7 October, which questions remained pending, which older replies had been superseded by Draft 4, and why the remaining uncertainty did or did not alter the outcome. Such a statement would not guarantee agreement. It would give an appeal a stable target.

The absence of that complete receipt is consequential because later readers otherwise face a false choice. They may assume that the older objections condemned Draft 4, or assume that the deletion necessarily cured everything. Both positions exceed the sources. The proper finding from the sealed record is narrower: the public evidence did not close the version-to-version loop. Any firmer conclusion would require additional, version-specific material.