The pool is the policy: AFRINIC’s recovered-space choice

Summary

AFPUB-2026-IPv4-001-DRAFT02 remained a proposal on 9 August 2026. Current AFRINIC policy already puts all IPv4 space managed by AFRINIC under the applicable Soft Landing exhaustion phase, whatever pool it originally came from.
AFRINIC is the bookkeeper of the free pool, not its sovereign owner. It may verify that recovered space is genuinely under its control, protect uniqueness and record accuracy, clean the prefixes, preserve network continuity, and allocate genuinely unallocated space under narrow, objective queue criteria. It may not convert scarcity, need, region, compliance or community procedure into a general power over operator capital.
Draft 2 would add a single available-space category, an open-ended power to shorten quarantine, priority for Soft Landing over inconsistent allocation criteria, and a mechanism for releasing the clean final-/8 /12 and refilling it with recovered addresses, including addresses still in quarantine. Those are substantive powers, not clerical clarifications.
Priority would still be shaped by a /22 maximum for each allocation or assignment, 90 per cent utilisation for most additional requests, and no explicit limit on request frequency. The evidence does not show that repeat applicants acted abusively. It does show that a nominally equal per-request rule rewards the ability to deploy, document and re-enter the queue quickly.
The proposal’s statement that AFRINIC had recovered more than three million addresses is an attributed official claim, not a prefix-level inventory. Recovery basis, disputes, quarantine, cleanup, fragmentation and allocable yield remain unreconciled. The draft should not advance as a bundled priority architecture. Bookkeeping required for safe allocation must be separated from discretionary governance that AFRINIC has no mandate to exercise.

Analysis

The bookkeeper’s exact job

AFPUB-2026-IPv4-001-DRAFT02 asks a deceptively simple question: what should happen when IPv4 addresses return to AFRINIC while the ordinary free pool is approaching exhaustion? The answer begins with a boundary that cannot be negotiated away by scarcity. AFRINIC is a registry. Its legitimate common-layer work is narrow: preserve uniqueness, maintain accurate records, verify control, isolate conflicts, retain security-relevant state, clean returned prefixes, record transfers and status changes, and protect the continuity of running networks.

Where a genuinely unallocated free pool still exists, it may also distribute that finite pool under objective criteria established in advance.

That limited allocation function does not make AFRINIC the owner of the pool in a sovereign sense. It does not convert a service region into territory or a mailing-list process into a legislature. It does not allow the registry to impose moral judgments about which commercial model deserves an address, to punish a resource holder through the ledger, or to extend a free-pool queue rule into continuing authority over addresses after they have entered operating networks. Allocation from an unallocated pool is a narrow administrative exception grounded in the need to choose among simultaneous claims on finite unissued supply.

It is not a constitutional source of power.

The distinction matters because Draft 2 is written at the point where bookkeeping turns into distribution. A prefix described as recovered is not yet ordinary supply merely because the registry uses that label. AFRINIC must be able to show the basis on which it treats the prefix as under its control. It must preserve any dispute state. It must prevent duplicate registration. It must record the block’s source and present condition. It must allow stale routing, routing-registry and deny-list history to settle or be corrected.

It must protect the next recipient from inheriting an address range whose operational residue has been hidden by an accounting change.

Only after those functions are complete can a recovered prefix become genuinely unallocated space available for a new issuance. At that point, AFRINIC may run a queue. The criteria must be objective, narrow, public and capable of consistent application. They may establish application completeness, a per-request quantity, an order of processing, and verifiable conditions relevant to allocating the remaining free stock without duplication or immediate waste.

They may not smuggle into the queue a claim that the registry owns regional IPv4 capital, that “the community” owns it, or that institutional discretion is legitimate merely because scarcity makes the decision consequential.

Draft 2 crosses this boundary in four connected moves. It would commingle every returned or recovered resource into one available-space concept regardless of original pool. It would allow the twelve-month quarantine to be shortened when available space falls below a /13 or for other operational reasons, without a normative minimum for the residual exception. It would make Soft Landing prevail over inconsistent allocation criteria elsewhere in the Consolidated Policy Manual. It would permit the protected final-/8 /12 to be released and refilled with recovered addresses, including addresses still in quarantine.

Each move alters who can obtain scarce capacity, when they can obtain it and which risks are carried by the recipient or by future applicants. Together they create a priority architecture. Calling it clarification does not make it clerical. The registry’s power must be judged by what the text lets it do, not by the administrative vocabulary used to describe it.

The proposal was not policy at the cutoff

Draft 2 was submitted by Jordi Palet Martinez on 16 June 2026. The official detail page contained an AFRINIC impact assessment dated 19 June. The proposal was presented at the AFRINIC-37 Public Policy Meeting in Nairobi on 24 June. A later co-chair notice said the meeting had reached rough consensus and opened Last Call through 31 July at 23:59 UTC. Objections and responses continued through the end of that period and immediately after it.

Those are facts about procedure. They are not adoption. On 9 August 2026, AFRINIC’s current-proposals index still listed Draft 2 as Under Discussion. The reviewed record contained no final co-chair consensus determination, Board ratification, implementation notice or CPM incorporation. The detail page and archive preserved the completed Last Call stage, but meeting rough consensus was not final consensus and was not a grant of current eligibility.

The institutional sequence must stay exact. The Policy Development Working Group discusses proposals and develops a consensus record. Co-chairs guide that process, evaluate Last Call and determine whether final consensus exists. If they determine that it does, they submit the draft for Board ratification. Staff implements adopted text. The Appeal Committee can examine whether a co-chair action complied with the policy-development process; it does not replace the working group’s merits judgment with its own.

AFRINIC describes this process as open, transparent and bottom-up. That is AFRINIC’s own account of its procedure. The archive shows that entities did engage: they offered support, conditions, objections, comparisons and implementation questions. Procedure can improve the information available to a decision. It cannot create sovereignty. Attendance is not mandate, rough consensus is not ownership, and Board ratification would not transform a private recordkeeper into a state. Even a properly adopted policy must remain inside the registry’s legitimate function.

That boundary is not an argument that the current PDP is void. It is a rule about scope. A procedurally sound decision can allocate a genuinely unallocated pool under objective criteria. It cannot acquire a power the institution never possessed merely because every procedural step was completed. The difference between procedural validity and substantive mandate is the difference between a queue rule and capital control.

Current policy already supplies the Phase 2 baseline

The strongest textual reason to resist the bundle is that current policy is not silent. CPM 5.4.3 already applies the Exhaustion Phase rules to all IPv4 resources allocated, assigned or otherwise managed by AFRINIC, whether or not a resource came from the final /8. Alain Aina raised that point in the archived Last Call discussion. The legal effect of the provision on every recovered-space question was debated, but its text establishes that non-final-/8 origin does not by itself remove managed space from the exhaustion framework.

Current Phase 2 rules set a minimum allocation or assignment of /24 and a maximum of /22. A /24 contains 256 addresses; a /22 contains 1,024. CPM 5.4.4 states no explicit limit on how many times an organisation may request additional IPv4 during the Exhaustion Period. CPM 5.4.5 uses an eight-month planning horizon. CPM 5.4.6.1 generally requires 90 per cent utilisation of all previous allocations or assignments before an additional request, with an exception for the first request by new LIRs and End Users.

CPM 5.2.4 says public-address policies and practices apply fairly and equitably to existing and potential members regardless of location, nationality, size or other factors.

These provisions amount to a queue architecture for unallocated supply. They place a ceiling on the size of one issuance, specify a deployment horizon, require documented use for most repeat access and state a general equality principle. Such rules can be legitimate only while they remain attached to genuinely unallocated pool distribution and are applied objectively. They do not create a continuing registry claim over the commercial life of an address after issuance. They do not make “need” a universal morality. They do not authorise punishment through registration.

Current policy also contains a rule for the final-/8 /12. CPM 5.4.7.1 preserves it for unforeseen future needs. CPM 5.4.7.2 already says that, if the /12 remains unused when the other remaining available space has been allocated, it returns to the Phase 2 pool. Draft 2 therefore does not invent every possibility of reserve release. It changes when release may happen and what may stand behind the reserve after release.

That is why the additional provisions cannot be merged into a single claim of certainty. Draft 2 would declare each exhaustion phase final and irreversible while it remained operative. It would erase original-pool distinctions in the available category. It would create an express quarantine-shortening power tied partly to a threshold and partly to unspecified operational reasons. It would give Soft Landing priority over inconsistent allocation criteria. It would allow the reserve to be replenished with addresses that are not yet through quarantine. It would delete CPM 5.5.1.2.1, 5.5.1.4.1 and 5.6.3.

It would shorten the 50 per cent utilisation period in 5.5.1.9 from twelve months to eight months and shorten the SAW period and refresh cycle in 5.5.1.13.3.2 on the same basis.

The current Phase 2 baseline may justify a short declarative rule that clean, genuinely unallocated recovered space remains eligible for Phase 2 allocation. It does not justify the rest. Universal commingling, an open exception, reserve substitution, supremacy and changed obligations need independent authority and independent necessity. The bookkeeper cannot obtain them by attaching them to a proposition the CPM already states.

Recovered is a status, not a title

Draft 2’s problem statement says AFRINIC had recovered more than three million IPv4 addresses. That is an official claim attributed to the proposal. It is not an audited prefix list in the reviewed public record, and it cannot be restated as though more than three million clean, undisputed, contiguous and allocable addresses existed on the cutoff date.

The unknowns are central. The public material reviewed for the sealed record did not identify the exact prefixes composing the aggregate. It did not reconcile original pools, recovery grounds, continuing disputes, quarantine start and completion, fragmentation, routing history, cleanup progress or allocable yield. It did not show whether the promised Secretariat information on the recovery process appeared during Last Call in a clearly identifiable prefix-level disclosure.

The vocabulary has to remain disciplined. A block may be described by AFRINIC as recovered while it is still in quarantine. It may be controlled but disputed. It may have completed one cleanup check but retain other operational residue. It may be reserved as a replacement quantity rather than available for immediate delegation. It may be clean but fragmented in a way that affects which request it can satisfy. Recovered, disputed, quarantined, clean, reserved and available are different ledger states.

The 2021 AFRINIC WHOIS audit demonstrates the distinction without resolving the 2026 figure. That historical report said 1,060,864 addresses had been reclaimed and placed in twelve-month quarantine. It separately reported 1,310,720 addresses not yet reclaimed after due diligence and identified other legacy-address categories. Those figures cannot be added mechanically to produce a current allocable inventory. They show why provenance and state must survive any pool classification.

The registry’s role here is both narrower and more demanding than ownership. It must tell the truth about state. It must not use a recovery label to claim sovereign title, and it must not erase a dispute merely by changing a database field. It may record that a prefix is held in a recovery process, quarantine it, protect against duplicate assignment, preserve evidence and move it to available status when the objective conditions for safe reissuance are satisfied. The ledger describes that sequence. It does not create whatever authority the registry would find convenient at the next step.

Commingling is therefore not an innocent data-model choice. If the public can no longer distinguish which prefixes were recovered, why they were recovered, whether they completed quarantine and how they were cleaned, the ledger stops describing reality at the moment the new allocation begins. The loss of provenance also hides whether a future policy still has an unissued recovered stock available to it. A single eligibility rule need not require a single opaque inventory. AFRINIC can apply Phase 2 to clean free-pool space while continuing to preserve source and state.

The apparent equality of a /22 cap

Draft 2 would leave the existing Phase 2 request machinery at the centre of access. The /22 maximum limits one allocation or assignment to 1,024 addresses. It does not establish an annual limit for one member. With no explicit request-count ceiling in current CPM 5.4.4, an organisation may deploy an allocation, satisfy the utilisation threshold, prepare a new complete request and return to the pool.

The proposal attributes to AFRINIC staff the observation that several Resource Members obtained more than eight /22s during one calendar year through multiple requests of up to /22. Eight /22s equal 8,192 addresses, so the reported threshold is above 8,192 for each member included in that observation. The proposal page did not publish the number or identities of those members, the intervals between their applications, their utilisation files, their business purposes, their countries, denials or their share of total supply.

That evidentiary gap prohibits a claim of abuse. Repeated requests were allowed by the stated rule. Rapid utilisation can reflect real customer growth and network deployment. Scarcity is not proof of hoarding. A registry may reject fraud, a duplicate claim or an application that fails objective free-pool criteria; it may not convert a lawful request pattern into moral wrongdoing after the fact.

The observation nevertheless shows why the queue cannot be described as equal simply because every request has the same maximum. The economically relevant unit is not one application but a member’s access across time. Deployment speed, administrative staffing, familiarity with Hostmaster review, the ability to document 90 per cent utilisation and the timing of a complete new submission affect how quickly an applicant becomes eligible again. The cap constrains transaction size. It leaves the repeat option open.

This is the poverty penalty built into discretion and procedure. A formal need rule does not neutralise unequal administrative capacity. Larger or better-staffed operators can often absorb delay and repeated review more easily. The sealed evidence contains no applicant-level data that measures the result, so the magnitude remains unknown. The absence of measurement, however, does not authorise the registry to correct suspected inequality through another layer of discretion. It requires an objective queue and transparent reporting.

AFRINIC could publish anonymised counts of first and repeat requests, their size distribution, time between successful requests, review duration, denials or deferrals by stated reason, and the proportion of available supply issued through repeat eligibility. Such reporting is part of truthful bookkeeping. It would not entitle staff to choose a socially preferred recipient outside the published rule. Fairness cannot mean an unpublished discretionary adjustment. It must mean that similarly situated complete requests meet the same rule and that the cumulative output is visible.

For operators, access has an economic effect whether the registry calls the addresses assets or resources. An administrative-price allocation can substitute for buying or leasing scarce IPv4. A delay can force a network to seek capacity elsewhere. A reused prefix can impose cleanup and reputation costs on a recipient if stale operational history survives. These consequences make precision necessary. They do not make the registry sovereign.

Three stocks and no honest combined runway

Draft 2 reported an AFRINIC staff snapshot of 783,872 addresses then available and an estimated two years to depletion at the then-current delegation rate. The count equals 3,062 /24s, or 765.5 /22 equivalents. Dividing by approximately two years yields an implied rate of 391,936 addresses a year, 382.75 /22 equivalents a year or about 31.90 /22 equivalents a month. This arithmetic exposes an assumption in the estimate; it does not validate a stable future rate.

At the 9 August cutoff, AFRINIC’s statistics portal displayed 2,562 available /24s, equal to 655,872 addresses. The dashboard showed no visible precise data-as-of timestamp. The difference between the proposal snapshot and the displayed cutoff count was 128,000 addresses, exactly 500 /24s. The evidence does not permit that difference to be called pure allocation burn. Allocations, reservations, reclassification, update timing or category semantics may each account for some part.

The final-/8 /12 contains exactly 1,048,576 addresses, or 1,024 /22 equivalents. The proposal attributes to staff an estimate of approximately three additional years to deplete it after the then-current available stock. That implies about 349,525 addresses, or 341.33 /22 equivalents, per year. It is a forecast assumption, not a guarantee. Adding the June snapshot and the /12 produces 1,832,448 addresses as exact arithmetic, but not a single reliable runway because the proposal describes sequential stages whose demand, fragmentation and approval patterns may differ.

The more-than-three-million recovered claim is another stock again. More than three million is strictly more than 2,929.6875 /22 equivalents. At the two rates implied by the staff estimates, three million would illustrate about 7.65 or 8.58 years. Those are sensitivity scenarios only. They assume that the lower-bound numerator is usable, that allocation rates remain constant and that quarantine, disputes, fragmentation, cleanup loss and policy reservation have no effect. None of those assumptions was established.

An illustrative division of the cutoff dashboard figure by the older implied annual rate produces about 1.67 years. That is not a forecast either. It assumes the category displayed in August is comparable to the June snapshot and that the earlier implied rate continues. Presenting it as a prediction would turn arithmetic into false evidence.

The correct ledger keeps at least three categories separate: currently available space, protected reserve and recovered space at each state. A number displayed in one category must not be silently added to another. The registry may allocate available free space. It may preserve a reserve under its stated rule. It may process recovered prefixes toward availability. It cannot make all three equivalent by declaration.

The strongest argument supports only the narrow function

The strongest case for Draft 2 is immediate continuity. African networks with valid capacity requirements continue to serve customers while the ordinary pool shrinks. Current CPM 5.4.3 already points managed space into Phase 2. Express language could stop repeated arguments that returned supply reopens Phase 1. A twelve-month default quarantine reflects the official concern that routing-registry entries, routing history and blocklist records may disadvantage a new recipient if space is released too quickly.

The reserve bridge also has an intelligible operational purpose. Rather than issue a newly recovered prefix before cleanup is complete, AFRINIC could give an applicant clean capacity from the untouched /12. A recovered quantity would be booked behind the reserve while quarantine proceeds. On the author’s account, this prevents valid pending applications from waiting and preserves a nominal /12 quantity. The author further argued that “final and irreversible” describes the phase while it is active, not a permanent ban on later CPM amendment.

Soft Landing supremacy is defended as a way to remove conflicts in the manual. The two twelve-to-eight-month changes are defended as alignment with the existing eight-month Soft Landing horizon. Staff’s impact assessment reported no required change to WHOIS, RDAP, MyAFRINIC, NetSuite, NMRP or RPKI, while proposing an automated cleanup and monitoring tool and estimating six months for implementation. A competing multi-pool architecture could require more classification and take longer to settle.

That is the contrary case in its strongest form. It proves that a narrow clerical problem exists. It does not prove the legitimacy of the bundled overreach.

Continuity supports preserving accurate registration and allocating clean unissued space without avoidable delay. It does not support an unrestricted “other operational reasons” power. Avoiding a return to Phase 1 supports a short phase rule. It does not support erasing provenance. Administrative coherence may support an enumerated conflict rule. It does not support a blanket supremacy clause whose full downstream effect has not been identified. A clean bridge may keep a complete request moving. It does not make quarantined fragments operationally equivalent to an untouched /12.

An eight-month horizon in one clause does not prove that existing twelve-month obligations may be changed without a transition rule.

Nor does the possibility of later amendment restore the inventory choice consumed today. A later PDWG can rewrite the CPM. It cannot recreate an unissued recovered prefix after that prefix has been delegated and embedded in a running network without disturbing the recipient or waiting for another recovery. Legal optionality remains; physical optionality disappears one allocation at a time.

The doctrine does not ask for kinder exercise of authority that AFRINIC lacks. It separates what the bookkeeper may do from what it may not do. AFRINIC may keep the free-pool queue moving. It may clean a prefix and protect the next user. It may maintain the unique and accurate ledger. It may not use those legitimate tasks as consideration for a broader discretion over scarcity. Safeguards do not convert capital control into coordination. The unauthorised power must be removed from the text.

Quarantine is a technical boundary, not a discretion reservoir

Draft 2 sets a default twelve-month quarantine and allows shortening when available space is below a /13, equal to 524,288 addresses, or for other operational reasons. Staff described a possible reduction to six months. The normative text contains no six-month floor. Tshepo Masuku put the gap on the RPD record, while the author treated quarantine duration as an implementation matter.

The text controls. A staff description cannot narrow a power that the normative clause leaves open. If “other operational reasons” has no enumerated conditions and no minimum, it gives the office that classifies and releases the prefix an undefined ability to convert uncertain inventory into economically valuable supply. That is discretionary scarcity governance. It is outside the bookkeeper’s mandate.

This conclusion does not require denying technical judgment. Staff must inspect routing and registration residue. Staff must perform cleanup. Staff may determine whether a stated deterministic exit test has been met. Technical application is not the same as policy discretion. The rule should state the fixed minimum or the objective evidence-based conditions under which a prefix is safe for reissue. If the condition cannot be specified with enough precision to constrain ordinary decisions, the prefix remains in quarantine until the default period expires.

The downstream risk makes the boundary necessary. A new recipient may face stale announcements, routing-registry entities, deny-list history, failed customer traffic and due-diligence work. No failure rate was quantified in the reviewed evidence, so it would be wrong to call every shortened quarantine dirty or harmful. It would be equally wrong to infer no operational effect from staff’s no-core-system-change table. Registry software can operate exactly as designed while the recipient bears the cost of a prematurely released prefix.

The institution that makes the acceleration decision does not carry every downstream loss. This is the familiar power-liability fracture: consequence-heavy control paired with thin exposure to the consequences. Publishing reasons after the fact does not cure a power that should not exist in open form. Auditability is required for legitimate bookkeeping, but auditability is not a sacrament that blesses discretion. The cure is an objective boundary and no residual policy power.

The /12 bridge is not a preserved reserve

The reserve mechanism exchanges time and quality. At the start, AFRINIC holds an untouched final-/8 /12 for unforeseen future need, while recovered prefixes sit in different states. Under staff’s intended sequence, if ordinary available space cannot satisfy a request, clean reserve capacity can be released. Draft 2 allows recovered prefixes to replenish the /12 quantity even while they remain quarantined.

The present applicant receives clean addresses now. The reserve receives a claim on capacity that may become usable later. Nominal count can remain /12 while operational quality, contiguity, history and time-to-use change. An aggregation of recovered prefixes with different quarantine dates is not the same present entity as one untouched /12. The proposal’s general cleanup sequence and its refill clause can coexist only if refill means an accounting reservation, not a claim that the replacement prefixes are already available.

AFRINIC may record such an accounting state truthfully. It may not call it preservation of the reserve in a way that conceals the changed entity. A truthful ledger would identify clean reserve prefixes released, recovered prefixes booked behind them, the quarantine state of each replacement and the point at which each replacement became actually usable. Until then, the institution has exchanged a present operational option for a future conditional one.

More fundamentally, the original purpose of the /12 cannot be changed by treating its size as a fungible column. Current CPM reserves the block for unforeseen future need and already describes an eventual return after other available space is allocated. Draft 2 would turn the reserve into a liquidity bridge once recovery begins. That is a new priority decision. It cannot be justified as mere accuracy or continuity because it decides that current queued applicants outrank the option held for an unforeseen use.

There may be a point at which the adopted rule objectively releases the /12. AFRINIC may execute that rule. It may not redesign the purpose through internal sequencing. Any replacement arrangement must be established expressly, and quarantined space must never be represented as clean reserve capacity. The ledger may record a future claim. It may not manufacture present equivalence.

One pool removes information that the ledger must keep

The competing AFPUB-2026-IPv4-003-DRAFT01, the Dynamic IPv4 Pools Exhaustion Management Framework, proposed separate traceable pools, a holding state, a pre-Soft-Landing pool, source tracing, anti-aggregation provisions and a later post-exhaustion queue. It was also Under Discussion at the cutoff and was not an adopted fallback. No comparative depletion or implementation model proved it superior.

The alternative matters because it demonstrates that Draft 2 is not the only imaginable rule. It also carries choices that need independent scrutiny. Larger possible delegations from a pre-pool, new needs tests, anti-aggregation and a later queue can themselves thicken governance. A more elaborate architecture is not doctrinally sound merely because it preserves more categories.

The valid comparison is narrower. A single allocation pool can use one familiar set of objective Phase 2 criteria. That operational simplicity does not require erasing source and state from the ledger. Separate bookkeeping can coexist with common eligibility after a prefix is clean. Conversely, preserving multiple allocation entitlements based on provenance could create new discretionary judgments that go beyond uniqueness and safe free-pool administration.

The correct design therefore separates the state ledger from the issuance queue. Every recovered prefix retains prefix-level provenance, control basis, dispute status, quarantine state, cleanup state, reservation state and allocation status. Only genuinely unallocated, clean prefixes enter the issuance queue. The queue applies narrow objective rules. After issuance, the registry records the new control state and protects continuity. It does not acquire an ongoing claim to dictate the holder’s geography, commercial model or moral worth.

This structure preserves the part of Draft 2 that is administratively necessary without accepting its sovereignty claim. It also avoids pretending that the Dynamic Framework’s additional eligibility choices are automatically legitimate. The question is not one sovereign pool versus another sovereign pool. The question is whether the registry can perform thin, truthful bookkeeping while a finite unallocated stock is issued under criteria that do not expand into governance.

What Draft 2 must lose

Draft 2 should not advance as the bundle published at the cutoff. Its legitimate core is smaller than its operative reach.

A valid narrow rule could state that a recovered prefix, after verified control, preserved dispute status, completed cleanup and objective quarantine exit, becomes unallocated space eligible under the current Phase 2 issuance criteria. It could preserve the current phase against automatic reopening when addresses return. It could correct exact textual conflicts necessary to make that issuance rule deterministic. It could require the registry to maintain accurate prefix state and keep the unique holder record current.

The rest must be removed unless it is separately shown to be indispensable to a technical invariant. Original-pool provenance may not be erased from the ledger. “Other operational reasons” may not operate as an unbounded exception. Quarantined prefixes may not be described as present clean reserve capacity. A blanket supremacy clause may not displace unidentified obligations. Changes to utilisation and SAW periods may not ride inside a recovered-space measure without separate justification and prospective treatment. Neither regional need nor community preference may become a title theory over the space.

This is not a request for more elaborate safeguards around the same powers. Published criteria, reasons and appeal can make a legitimate queue easier to audit. They cannot legitimise punishment, capital control or sovereign discretion. When a clause exceeds the registry function, the proper remedy is deletion, not a promise of careful use.

The resulting system is demanding precisely because it is narrow. AFRINIC must know what it controls and why. It must protect against duplicate use. It must preserve accurate public records and security-relevant assertions. It must clean recovered prefixes rather than export hidden residue to a new holder. It must keep running networks continuous. It must apply the unallocated free-pool rule objectively. It must record what happens.

It may not rule.

The decision at the cutoff

By 9 August 2026, the evidence established an active scarcity problem but not the inventory claimed by the architecture. The proposal attributed more than three million recovered addresses to AFRINIC, yet no public prefix-level reconciliation showed how much was undisputed, how much had completed quarantine, how much was clean, how much was fragmented, how much was booked as reserve replacement or how much was ready for allocation. The dashboard displayed 655,872 available addresses without a visible precise snapshot time. The proposal’s older 783,872 snapshot could not be reconciled to it from the public record.

The evidence also established a current policy baseline. CPM 5.4.3 already applies Exhaustion Phase rules to all managed IPv4 space. CPM 5.4.7.2 already defines a future release path for the untouched /12. Draft 2 therefore was not necessary to prevent all use of recovered space or to invent all reserve release. Its real additions were commingling, discretion, substitution, supremacy, asserted irreversibility and timing changes.

The strongest pro-Draft 2 argument was that valid applicants should not wait for a perfect architecture while clean supply runs down. That urgency supports prompt, accurate clerical work. It supports a deterministic interim rule. It does not support institutional self-enlargement. Scarcity narrows the acceptable mandate because mistakes become more consequential; it does not turn the clerk into the owner of the scarce entity.

The decision should therefore have been to reject the bundle and separate the functions. Preserve a public prefix-state ledger. Apply current Phase 2 only after a recovered prefix becomes clean, undisputed unallocated space under objective conditions. Retain the default quarantine absent a deterministic early-exit rule. Keep quarantined replacement claims distinct from usable reserve capacity. Examine every unrelated timing amendment separately and prospectively. Publish cumulative queue outcomes without granting staff power to engineer a preferred distribution outside the rule.

That conclusion does not condemn recovered addresses to permanent idleness. It gives them a lawful administrative path back into use. Nor does it deny AFRINIC an important role. The registry’s thin role is indispensable while operators rely on its records: it protects uniqueness, accuracy and continuity and makes state transitions legible. What it denies is the alchemy by which importance becomes sovereignty.

The recovered pool is not regional political property. It is not community property. AFRINIC’s record does not create ownership by AFRINIC. A valid free-pool allocation rule is a queue discipline, not a permanent theory of operator obedience. The ledger must remain a ledger from recovery through reissue and after the prefix enters a running network.

The uncertainties that remain part of the record

Ten unresolved questions prevent the draft’s architecture from being treated as a settled description of supply. The exact prefixes behind the more-than-three-million official claim are unknown, together with their original pools, recovery basis, dispute condition, quarantine completion, fragmentation, routing cleanup and allocable yield. The June 783,872 snapshot and the cutoff display of 655,872 have not been reconciled by dated allocations, reservations, reclassifications or category definitions. No final co-chair consensus decision, Board ratification or implementation record was located by the cutoff.

The minimum shortened quarantine is also unknown because staff described six months while the normative language set no such floor. The exact accounting, sequence, security condition, public reporting and use conditions for /12 replacement prefixes still in quarantine were not published. The member count, identities, timing, utilisation and purpose behind the reported more-than-eight-/22 pattern remain undisclosed, so no concentration or abuse conclusion can be drawn.

Draft 2 did not say how the two eight-month changes would apply to assignments, SAWs or renewals already approved on twelve-month terms. It did not enumerate every CPM conflict that the broad supremacy clause could displace. The searched record did not locate a clearly identifiable prefix-level Secretariat disclosure of the recovery process promised during Last Call. Finally, the two- and three-year official estimates were not supported in the reviewed public record by the underlying daily series, demand funnel, pending-request count, approval rate, fragmentation, recovery yield or sensitivity model.

These are not gaps for AFRINIC discretion to fill privately. They are boundaries on what the proposal can truthfully claim.