Summary

  • The IANA Numbering Services Review Committee's 2025 report, dated 30 March 2026, concludes that the operator's performance was acceptable, records reverse DNS meeting its service target 80% of the time against 100% everywhere else, and prints no tolerance, weighting or aggregation rule that produces the verdict.
  • The instruments that carry the decision rules print the same fallback threshold on two different bases, and the report that carries the findings restates neither the threshold nor the quorum nor the split between voting and non-voting members.

The sentence is one line long. "We, the RC, conclude that the performance of the IANA Numbering Services Operator related to delivery of the IANA Numbering Services and performance under the SLA in 2025 was acceptable." It sits in the 2025 report of the IANA Numbering Services Review Committee, a document dated 30 March 2026 covering 1 January to 31 December 2025.

Two paragraphs away, the same report prints what the word rests on: the service level agreement was met 100% of the time in all cases except reverse DNS, where it was met 80% of the time, with a June spike in DNS propagation time observed. In the sections reviewed, the report prints no tolerance for a family below target, no weighting between service families and no aggregation rule. Nothing on the face of the document explains how an 80% result in one family, sitting beside 100% in the others, becomes the word acceptable.

That is not a claim that the verdict is wrong. It is the narrower and more awkward point: a reader holding the report cannot rebuild the conclusion from it. The arithmetic that turns percentages into a verdict exists somewhere, or the word is doing work no calculation supports — and the record reviewed here does not let an outsider tell which.

The committee itself is fifteen seats. Two community appointees from each of the five RIR regions, who must not be RIR staff, plus one non-voting RIR staff member from each region: that is the charter's formula, and Operating Procedures v2.1 opens with the same one. The seats are not filled uniformly. The 2021 handbook records AFRINIC and ARIN seats going to elected NRO NC/ASO AC members, APNIC holding one community-elected and one APNIC EC-appointed member on two-year terms, LACNIC running an annual open call for one of two elected representatives with a Board-appointed staff member, and RIPE NCC filling its two seats with community-elected NRO NC members plus a RIPE NCC Board appointee — with terms listed as three years for AFRINIC, two for APNIC, three for ARIN, two for LACNIC and three for RIPE NCC. RIPE-813 ties RIPE's two seats to its NRO NC representatives, ending automatically on resignation or dismissal from that body. APNIC's own election page describes one seat elected by its community and one appointed by its Board, which is a labelling difference from the charter's "community appointees" and matters only if a reader tries to count how many of the ten voting seats were community-chosen. The charter itself was published after a consultation on a draft opened on 23 June 2015.

What those fifteen may do is narrow, and printed. The committee advises. It cannot commit the NRO EC to any recommendation or decision except where it holds express delegated authority; it must report concerns about the operator's performance, including failure or near-failure to meet SLA obligations, to that council at least once every calendar year by dates the council sets; it meets at least twice a year by teleconference open to the public; and it must consult the Internet Number Community before any final report or recommendation.

Who receives the word is printed too. The NRO Executive Council is one person from each RIR, appointed by each RIR's Board, deciding by consensus of all five. Each RIR appoints its committee members by a published method of its own choosing, sets and publishes their terms, and may remove or replace them on 30 days' notice to that same council, v2.1 says. The contract being reviewed is between ICANN and those five RIRs — approved for signing on 27 May 2016, signed 29 June 2016, effective 1 October 2016, renewing every five years absent six months' notice — and PTI performs ICANN's obligations as subcontractor with ICANN fully liable for the performance the committee grades.

Put those two facts side by side and nothing improper has happened. But the reviewers are appointed, and removable, by the counterparties to the contract, and their findings are received by a council built from those counterparties' appointees. The body checking delivery answers to the parties that bought it. That is a design the documents state plainly, and it raises rather than lowers the value of the one check that does not depend on the parties: whether a stranger can reconstruct the verdict from the papers.

The fallback threshold is where the papers begin disagreeing with each other. The charter sets it as supermajority approval of "eight out of 10 community members" if rough consensus cannot be reached. Operating Procedures v2.1, dated 11 December 2023 and replacing v2.0 of 10 October 2019 and v1.0 of 30 April 2018, sets it as 80% supermajority approval of eligible community members excluding abstentions. Eight of ten is 80% — that is my arithmetic, not the documents'. But excluding abstentions changes the base: the moment anyone abstains, the pool shrinks and the count required falls below eight. Two texts, one threshold, two denominators. I print both and crown neither.

Quorum moved as well. The 2021 handbook requires eight members including at least one from each region. v2.1 requires eight including at least one member from at least four regions and five voting members. The NRO's programme page calls v2.1 current, so the difference is dated rather than ambiguous — and the later text permits a quorate session with one region absent. No minute, quorum call or published determination from any 2025 meeting was obtained in this review, so what was actually satisfied on the day the report was settled is not established here.

The 2025 report then does something neither instrument does: it prints its own composition. "The RC is comprised of three representatives from each RIR region," followed by thirteen names — three each for APNIC, ARIN, LACNIC and RIPE NCC, and for AFRINIC a single name labelled RIR Staff. Fifteen designed seats against thirteen printed names leaves two unaccounted for; the subtraction is mine, and the report does not say which two. Nor does the reviewed text say whether the two AFRINIC community seats were vacant, unfilled for the reporting period, or lost in the extraction I read.

The report prints no voting and non-voting split either, so a reader cannot tell from it which of those thirteen names decide.

This is not one year's omission. The 2023 report and the 2024 report each conclude acceptable with the SLA met 100% of the time in all cases, and the report index records no comments for the 2023, 2024 and 2025 cycles; the 2025 landing page repeats the nil return for the comment period that ran 4 February to 6 March 2026. Document dates and index dates diverge in two of those years — the 2023 report's text dated 29 February 2024 against publication listed as 5 March 2024, the 2024 report dated 13 March 2025 against 16 March 2025 — a signing-versus-publication difference rather than a contradiction. Silence in a comment period is not assent, and the record does not tell a reader whether it was satisfaction or inattention.

Three printing decisions would make the record auditable from itself. Restate in the report the quorum satisfied, the voting split, and the rule converting service-level percentages into a verdict, including whether reverse DNS at 80% falls inside or outside whatever tolerance applies. Reconcile the fixed count with the percentage, or say explicitly which governs. And print fifteen names, or say why two are missing. None of those is a policy change. Each would let an outsider reproduce an outcome that at present rests on trust.

The contract carries escalation and dispute-resolution mechanisms; no 2025 record of either was obtained in this review. So the annual report is, in practice, the public account of whether the numbering services were delivered as promised — and the one step that would let anyone check the word acceptable is the step no reviewed document takes. The verdict is not the problem. The missing bridge is.