Summary

  • ICP-2, accepted by the ICANN Board on 4 June 2001, is written as an entry test: ten unranked criteria it calls "essential preconditions", with approval of new registries assigned to ICANN. The retrieved text carries no provision for withdrawal, suspension or derecognition, and no successor-transition mechanism.
  • The ICANN Board's Secretary's Notice of 24 December 2024 prints that gap itself — formal derecognition and transition to a successor registry "are not covered" — while the draft that would cover them, labelled "(FINAL DRAFT)" and not adopted, sets two two-thirds gates over two different populations and lets a registry's own members open the procedure.

Read ICP-2 the way a stranger would have to. It is titled "Criteria for Establishment of New Regional Internet Registries", and the Address Supporting Organization's policy index and the ICANN-hosted copy agree on how it arrived: developed through the ASO with the assistance of APNIC, ARIN and RIPE NCC, recommended by the ASO Address Council, and accepted by the ICANN Board of Directors on 4 June 2001 "as a statement of essential requirements for the recognition of new Regional Internet Registries (RIRs)" (ICP-2 text; ASO policy index).

Who decides, in the document's own words: "According to the Address Supporting Organisation Memorandum of Understanding (ASO MoU) it is the responsibility of ICANN to approve the establishment of new RIRs." And what the ten criteria are for: "The existing RIRs agree that these criteria are essential preconditions for the approval of new RIRs, and that they provide ICANN with the minimum guidelines needed when evaluating applications for recognition of new RIRs."

Minimum guidelines, for evaluating applications, addressed to an organisation "proposing to establish a new RIR". ICP-2 also refuses to let anyone prefer one criterion over another: "The numbering which is used is not significant – all criteria are considered essential in an organisation which is proposing to establish a new RIR." Ten items, no ranking, each essential, every one pointed at an applicant.

The ten, in the order the retrieved text sets them out: a region of approximately continental size served by a single registry "established under one management and in one location"; "broad support of the LIRs (ISP community)", with "a very substantial majority of the ISPs in the region" prepared to support the new registry; bottom-up self-governance for local policy; neutrality and impartiality; technical expertise; adherence to global policies on conservation, aggregation and registration; a published activity plan; a funding model; record keeping, with "English is considered the official language of the registry system"; and confidentiality. The NRO's hub page states the scope in one line: ICP-2 "sets out the criteria for establishment of new RIRs" (NRO ICP-2 hub).

Now the gap. The retrieved ICP-2 text contains no provision for withdrawing, suspending or derecognising a registry already recognised, and no mechanism for handing a service region to a successor. On its own that is a thin claim — an absence in what this review could read.

It stops being thin because the institution that would have to act printed it.

On 24 December 2024, by Resolution 2024.12.24.01, adopted by unanimous consent without a meeting, the ICANN Board ratified the Implementation and Assessment Procedures for ICP-2 Compliance (Secretary's Notice; procedures, as ratified). The rationale of that act states: "Two of the main issues that are not covered within the Implementation and Assessment Procedures are (1) formal derecognition of an RIR; and (2) transition procedures to a successor RIR." It gives the reason in the same passage — that covering them would impose new obligations requiring policy development through the numbering communities — and records that "The RIRs agree that ICP-2 remains in force today", that the Board has used the 2001 text "twice" to recognise new registries, and that the ratified procedures "do not amend those obligations in any way".

Three sentences, one decision: entry is specified, exit is not, and the deciding body said so while ratifying the machinery that handles everything else. Arithmetic, mine and printed nowhere: 4 June 2001 to 24 December 2024 is 23 years, 6 months and 20 days.

Two boundaries belong beside that paragraph. What the compliance procedures themselves contain is a separate question, and this review did not open them: their existence, their ratification date and the Board's description of them are on the record, their sections are not quoted here.

The exit the 2001 text does not print now exists as a draft. The NRO's version table traces the lineage — the original "founding document of the RIR system, and is being updated through this process", a draft of 14 April 2025, Version 2 on 28 August 2025, and the ASO AC Recommended Draft on 1 September 2026 (NRO ICP-2 hub).

Version 3 is dated 1 September 2026 and labelled "(FINAL DRAFT)" (Version 3). The label is not an adoption. The ASO's announcement of the same date describes the text as the "ASO AC Recommended Draft RIR Governance Document", records that the NRO Number Council concluded its ICP-2 review, and states that further community-input opportunities will be announced by the NRO Executive Council (ASO announcement; NRO announcement). No reviewed page records adoption, and nothing here establishes that the draft will be adopted, amended or dropped.

Read as text, it restructures recognition into six criteria — Regional Qualification, Resource Holder Support, Community Support, Capability, Impact and Improvement — and then fences the recogniser: "ICANN shall have no power to Recognize a Candidate RIR unless it has received a Proposal to do so that is supported by at least two-thirds of the RIRs submitting Recognition Assessments." A registry whose service region or revenue would be materially affected may have a negative assessment set aside, but only after written notice, stated reasons, an opportunity to respond and fair consideration of the response.

Article 6 does the same for exits.

Derecognition requires material non-compliance with §4.1 that persists despite a reasonable opportunity to remedy; §6.2 presumes rehabilitation and requires a plan and a cure period before any final decision; §6.3 lets a proposal be initiated by any registry, by "a group of at least 25% of the subject RIR's total Members or 2,000 Members (whichever is lesser)", or by ICANN; and §6.7(c) provides: "ICANN shall have no power to Derecognize an RIR unless it has first received a Proposal to do so that is supported by the Derecognition Assessments of at least 2/3s of the RIRs, excluding the subject RIR."

Two two-thirds, two populations. Recognition divides by "the RIRs submitting Recognition Assessments" — those that chose to assess. Derecognition divides by "the RIRs, excluding the subject RIR" — a body that depends on who assessed not at all. Neither denominator appears as a figure on any page read for this piece (Version 3; summary of differences).

My arithmetic, kept off every printed figure: on a five-registry population, two-thirds is 3.33, so four registries; excluding the subject, two-thirds of four is 2.67, so three. The member trigger cannot be computed at all — 25% and 2,000 are printed, the member base is not — so its reach is unmeasurable from what this review could read.

The gate itself moved. Version 2, dated 28 August 2025, barred ICANN from acting without a proposal "approved by the RIRs" and set derecognition at unanimity of the other registries; it wrote the rehabilitative rule as a flat sentence: "A non-compliant RIR shall be Derecognized only as a last resort" (Version 2). Unanimity to two-thirds is a dated change, bracketed by 28 August 2025 and 1 September 2026. No reviewed page records who proposed it or why. That interval is 1 year, 8 months and 8 days — mine again.

Put the 2001 text beside the draft and the reallocation is plain, and describing it needs no motive. In 2001, approving the establishment of a registry is ICANN's responsibility. In the 2026 draft, ICANN has "no power" to recognise or derecognise without a qualifying proposal carrying peer assessments. A body that once applied ten criteria now receives a proposal and gives the reasons.

So the readable fact is narrower than a scandal and firmer than a reassurance. Entry is specified in prose a stranger can check. Exit is specified in fractions whose denominators are not printed. The trigger sits partly inside the registry whose status the procedure would decide. And the exit that would close the gap is still a draft — recommended by the Address Supporting Organization Address Council, published for comment, not adopted. What would change the reading is concrete: a page printing a member count beside the 25%/2,000 trigger, or an adopted instrument carrying the two-thirds gates into force with its denominators counted. Until then the entry test stands, the exit is a draft, and no adopted instrument prints the way back out.