Summary

  • Draft Policy ARIN-2026-2 would allow a subsequent IPv6 request when at least 75% of the prefix units defined in the organisation’s current 2-year network plan have been assigned or reallocated.
  • The observed numerator is not enough to reproduce the result. A plan revision can change the unit definition or planned-unit count and therefore change the denominator without changing deployed space.
  • The draft remains Under Discussion. The checked record does not show adoption, implementation, a staff and legal review, a real request or a measured denominator dispute.
  • A protected plan plus a compact calculation receipt can freeze the plan digest, effective window, unit definition, numerator, denominator, evidence cutoff, result and decision without exposing customers or internal topology.

Take a deliberately artificial organisation with sixty prefix units assigned or reallocated. Under plan A, which defines eighty units for the relevant two-year horizon, the arithmetic is 60 ÷ 80 = 75%. Under plan B, which defines one hundred units, it is 60 ÷ 100 = 60%. Plan A reaches the proposed gate; plan B does not.

This is not an ARIN case, forecast or allegation. The numbers are symbolic, chosen only to isolate the rule. The numerator is held constant. The denominator moves. That simple exercise reveals the documentary state hidden inside a percentage that otherwise looks as objective as a meter reading.

ARIN’s official page identifies ARIN-2026-2, “NRPM Section 6.5 Revision”, as a Draft Policy under discussion. Its current text is dated 4 May 2026; the history records its advancement to Draft Policy on 24 June. The proposal says the present Section 6.5 is complex, formula-driven and difficult to interpret consistently, and seeks to replace much of its sizing and utilisation machinery with planning-based evaluation.

That change is real at the level of design, but it does not abolish arithmetic. Proposed Section 6.5.4 would make additional IPv6 space available when four conditions are demonstrated. The first is the 75% test. The others concern deployment according to the previously submitted hierarchy, or a technical justification for architectural changes; a revised plan covering additional prefix-unit needs over a new five-year horizon; and continued aggregation, with contiguous additional space issued whenever possible.

Those are different propositions. Passing 75% is not the same as proving that a hierarchy was implemented. A technical justification for changing architecture is not a retrospective amendment to observed utilisation. A revised five-year projection is evidence about future demand, not the two-year denominator used to test the present request. Contiguous extension is a stated operational preference, not an entitlement produced by the calculation.

The word “current” carries the decision

The draft does not specify how a two-year plan receives a durable identity, when a revision becomes current, which version governs a request already being assessed, or whether the size of a “prefix unit” may change between versions. That absence does not prove that ARIN staff lacks an internal method. It means the published draft does not give a reader a reproducible selection rule.

The distinction matters because a plan is not an inventory. An inventory can show which resources were issued and which assignments or reallocations were recorded by a stated date. A plan also contains choices about future structure: typical unit size, hierarchy, sites or customers, and projected use. Those choices may legitimately change as a network changes. The problem is not revision. It is letting a later revision erase which documentary denominator governed an earlier decision.

The current NRPM provides the useful contrast. Version 2025.1, effective 3 March 2026, defines IPv6 utilisation in Section 2.16 as provider assignment units assigned divided by total provider assignment units. Current Section 6.5.3 then offers alternative subsequent-allocation predicates, including 75% of total address space, more than 90% of a serving site, or more than 90% allocated to serving sites. Those rules have their own interpretive questions, but their denominators are tied to holdings or serving-site structure. The proposed test instead points expressly to units defined by a current planning document.

Planning can be a better description of IPv6 architecture than blunt address consumption. Hierarchical deployment, aggregation and growth do not fit neatly into an IPv4-style scarcity gauge. Yet replacing a poor formula with a plan does not make denominator identity optional. It makes version control part of the policy result.

One numerator, two permissible readings

The symbolic comparison can be written as a decision test rather than a story about an applicant:

Decision-time input Plan A Plan B What changed
Assigned or reallocated prefix units at the evidence cutoff 60 60 Nothing
Planned prefix units in the two-year denominator 80 100 Plan version or scope
Calculation 60/80 60/100 Denominator only
Result 75% 60% Gate reached under A only

The example does not decide which plan ought to govern. “Current” might ultimately mean the version submitted with the request, the latest accepted version at an assessment cutoff, a rolling plan valid on the decision date or something else. Each design can be made legible. The draft does not choose among them in the checked text, so the article should not pretend to do so on ARIN’s behalf.

Nor does the example imply manipulation. A denominator might change because a new customer class uses /56 rather than /48, because an acquisition adds sites, because a hierarchy is redesigned for aggregation or because forecasts are honestly revised. Those may be reasons for change, not evidence of abuse. Accountability begins by preserving the change, not by assigning a motive to it.

Freeze the test, not the network

The minimum control is a decision-time snapshot. ARIN could retain the full network plan as protected application material and expose only a compact receipt. The receipt need not list customers, sites, internal segments or addresses. It needs enough identity and arithmetic to allow the organisation, reviewer and later policy auditor to know which test actually ran.

Receipt field Why it belongs
Draft or adopted policy and NRPM version Binds the decision to exact authority and wording
Request and relevant holding identity Defines the case and inventory context
Two-year plan identifier or digest Freezes the document called current
Plan effective time and assessment cutoff Makes the temporal selection rule visible
Prefix-unit definition States what one denominator unit means
Planned-unit denominator Exposes the base of the percentage
Assigned or reallocated numerator Records the observed count used
Exclusions and evidence date Fixes the calculation boundary
Percentage and gate result Preserves the actual arithmetic
Prior hierarchy evidence or change justification Keeps deployment consistency separate from the percentage
Revised five-year plan identity Separates future need from the current test
Contiguous-space availability and outcome Separates qualification from issuance geometry
Decision, notice, review and correction Attributes the result and makes it repairable

The full plan and evidence may remain confidential. A digest reveals no topology by itself. Public aggregation can be narrower still: counts of requests assessed by policy version, plan-selection method, pass or fail result, architectural-change branch and correction status. The aim is not to publish a customer ledger. It is to prevent the phrase “75% under the current plan” from becoming an untestable conclusion.

What the sources do not establish

The official page does not report an Advisory Council recommendation, Board adoption, implementation notice or completed staff and legal review for this draft. It does not show that an applicant submitted two plans, that ARIN selected one, that a result changed, that an allocation was granted or refused, or that routing or renumbering consequences followed.

The proposal’s statement that the revised model replaces formula-based and utilisation-based allocation mechanisms must also be read alongside its own proposed 75% condition. The model changes what is counted and why; it does not remove every quantitative gate. Likewise, the expectation that organisations begin using issued IPv6 space within twelve months is not evidence that any organisation complied or failed to comply.

The recommendation here is therefore procedural and prospective. If the draft advances, define the plan-selection rule and preserve its decision-time inputs. No finding about current staff practice or applicant conduct is required to see why that record would matter.

Sources