Summary
- The 2025 review concludes that IANA's numbering performance was acceptable and that the service level agreement was met 100% of the time in all cases except reverse DNS, where it was met 80% of the time.
- In the month the propagation record visibly missed its ten-minute ceiling, the published data reduces 130 implementations to six statistics: count, mean, shortest, longest, median and standard deviation. No transaction is named.
- The committee that produces the verdict is advisory in the charter's own words: it has no power to commit the registries' executive council to any recommendation or decision, except where expressly delegated.
- The annual question is a matrix the five registries write, and the only route from material failure to a remedy belongs to those same five.
On 4 February 2026 the IANA Numbering Services Review Committee opened a thirty-day comment window on its 2025 report. The window closed on 6 March and drew no comments. The report's conclusion runs to one judgement and one percentage: performance in 2025 was "acceptable", and the service level agreement was met 100% of the time in all cases except reverse DNS performance, where it was met 80% of the time.
That is the verdict. Now look at the month underneath it. The June 2025 performance report prints the propagation target as "≤ 10m (99.00%)" and, in the value column of the same row, 42.84m. Directly beneath, for a stated 130 implementations, it gives a mean of 20.10m, a shortest time of 0.01m, a longest of 44.20m, a median of 4.92m and a population standard deviation of 18.77m. No implementation, request or ticket is named.
My arithmetic, not the report's: at a ten-minute ceiling with 99.00% compliance over 130 implementations, at most one implementation could have exceeded the ceiling — 129 out of 130 is 99.23%, while two failures would be 98.46% and fail the target. Yet a block with a median of 4.92m, a maximum of 44.20m and a mean of 20.10m describes a fat tail, not a single outlier. The page also prints two means: 42.84m in the target row and 20.10m in the statistics block, with no caption in the extracted text saying which measures what. Either figure sits above a ten-minute ceiling. Both are above it, and neither carries a name.
It may be that the percentage and the durations measure different things; the page as extracted does not say so.
The committee's own account of that period is thinner still. Its 2025 report notes "a spike in June in DNS propagation time" and records reverse DNS at 80%. A reader of the assurance chain therefore learns that a spike occurred, in a named month, of unspecified size, affecting an unspecified number of transactions. The reviewers do not name them either, because the data set they cite — the monthly reports — does not contain them.
Allocation requests are the opposite case, and the contrast is the whole design. The June 2026 report itemises a single allocation request for 1 June: an AS Number, received at 10:56:42 from RIPE NCC, acknowledged at 17:17:11 on time within two business days, implemented at 19:54:42 on time within four business days and accurately implemented. Every step is timestamped, every interval is measured, and the accounting is genuinely transactional. What the row does not carry is a request, ticket or case number. The row's key is a date plus a request type, and the allocated resource prints as the placeholder "Implemented using resource(s)". In the same month, 126 reverse DNS implementations are reported as aggregates only.
With one request a month, a date and a type usually do pick out a single real event, and that is not a small thing. But it is not an identifier. Two AS number requests of the same type on the same day could not be told apart, and a reader qualifying a resource — the one check that would make the row evidence rather than a summary — finds a placeholder where the object should be.
The October 2026 report shows what the summary layer does when there is nothing to summarise. Its preamble states that individual requests "are expanded upon later in the report". That month had zero allocation requests and zero reverse DNS implementations. The allocation rows nevertheless print "Requests acknowledged on time (100%)", while the reverse DNS rows print a dash and a count of zero. One empty month is therefore rendered as perfect compliance on one metric and undefined on another, in the same table. Anyone auditing this record begins by asking which of those two conventions governs the headline.
The index page above all of it publishes no metrics at all: a delivery statement and a grid of monthly reports from 2017 to 2026. On 18 November 2026 two fetches of that page labelled a different month as latest, so even the word "latest" is not a stable fact about it.
Now the instrument. The operating procedures, version 2.1 of 11 December 2023, set the annual machinery: "At a time of the NRO EC's choosing, the RIRs will publish an IANA Numbering Services Review Matrix in January every year, summarizing the RIRs' review of the performance of the IANA numbering services in the last year." The committee forwards that matrix to the five communities for thirty days of comment and summarises the comments. The extraordinary route is narrower still: "In the event there is a requirement to complete an expedited review, the NRO EC chair will advise the Review Committee chairs of the requirement." The annual question is written by the five customers. The urgent question is opened by their chair.
The charter is candid about the body it creates. The committee's function is to advise and assist the registries' executive council in reviewing the service level of the numbering services; it must report concerns about the operator, "including observed failure or near-failure to meet SLA obligations", at least once each calendar year by dates the council sets. And then the sentence that decides what kind of thing this is: the committee "does not have any power to commit the NRO EC to any recommendation or decision made by it, except where and to the extent that it has express delegated authority from the NRO EC."
Its composition is a serious attempt at distance within that frame: fifteen members, two community appointees per registry region who may not be registry staff, plus one non-voting staff member per region. Each registry appoints by a method of its own choosing, sets its appointees' terms, and may replace them on thirty days' notice. Meetings are by teleconference, public unless confidentiality requires otherwise, at least twice a year; minutes go to a public mailing list and the NRO website; before any final report the committee must seek the community's feedback; decisions run on rough consensus, falling back to eight of the ten community members. The NRO's committee page publishes the roster and the document set, and its entry in this directory records the same fifteen-member body. In the charter text I read there is no channel at all for a party outside the council or a registry to raise a matter.
The service level agreement supplies the clocks and the lever. Article 4.3 requires acknowledgement within two business days, a request for further information within four, a registry update within four, a status report within ten where fulfilment is impossible, and allows an unsatisfied valid request after thirty business days to be asserted as a failure to perform, subject to Article 9. Article 6.2 requires monthly public performance metric reports, an annual numbering service survey and an annual security and systems audit. Article 8 lets the registries review the service and obliges the operator to cooperate. Article 9.2 gives the lever to precisely one set of hands: where the registries determine a material failure, they may invoke dispute resolution under Article 13, with termination under Article 10 to follow a determination.
The ICANN page fixes the agreement's dates — signed 29 June 2016, in effect 1 October 2016, board approval on 27 May 2016, automatic renewal every five years absent six months' notice — and states that the number community recommended forming this committee to advise the review. The NRO's register records that the operative text has moved: Amendment No. 1 of 12 November 2024 incorporated reverse resolution services. I did not read that amendment, so the article numbering quoted here may not be current, and I am not asserting what it changed beyond the title the register gives it.
Two divergences are worth keeping rather than tidying. The charter has the committee select both chair and vice-chair from its community appointees; version 2.1 has the chair elected annually and the vice-chair chosen by the chair from a different region. The charter sets the fallback at eight of ten community members; the procedures express it as 80% of eligible community members excluding abstentions — the same arithmetic in different words, and the words are different. The charter announcement says the council opened consultation on 23 June 2015 and published the final text on 5 November 2015 after it, and records no dated vote; the committee landing page says the charter was "accepted by the ICG", which no document in the set corroborates. Two statements about who adopted what, held apart because they are not the same statement.
What this adds to what was already published here is a reading of the instrument rather than an account of the body. On 18 November 2026 an exact-name search of this archive returned no rows; a broader search on IANA returned explainers on what the function is and how it relates to the registries, a strategic-plan webinar notice, a contingency-plan test report, an adjacent briefing on resource ownership, and a good deal of noise. None of it read the charter, the trigger, the powers or the published output. That is the delta claimed here.
The arrangement is not a fiction. There is a committee, a roster, minutes, a public list, an annual report, a thirty-day window, and a contract that measures acknowledgement in business days. What it is not is a route for the affected: the annual question is written by the five registries, the urgent question is opened by their chair, and the only lever that ends a failure is held by the five of them. Everyone else is left with a percentage — and in the one month that percentage mattered, it did not come with a name.
What would change that is not a matter of authority. A request identifier in the allocation rows costs one column. Publishing, each month, the number of implementations that exceeded the ceiling — beside the mean rather than instead of it — converts an average into a count somebody can falsify. Publishing an expedited review, when one is next convened, would show the extraordinary route working rather than merely existing. Each of those is a decision that the published record should be checkable, and each of them is available to the people who already write the report.
One limit on the reading, then, stated once: everything above comes from page and PDF extraction on 18 November 2026, not from a rendered browser. The monthly reports carry an expandable control I could not exercise, so I cannot exclude that the rendered page exposes a resource-level link the extracted text does not show. If it does, the identifier question narrows to one page and one control, and the finding is smaller than it looks tonight.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






