Summary

  • The 2016 numbering services agreement and its 12 November 2024 amendment state measurable, dated duties and a graduated remedy, but the remedy waits on a "materially established failure" finding, and the body that assesses performance has an advisory mandate.
  • The published record holds both halves at once: a June 2025 propagation row of 42.84 minutes against a ten-minute threshold, a 2025 review that logged 80% reverse-DNS compliance and still called performance acceptable, and an October 2026 report printing 100% acknowledgment over a month with no requests at all.

The promises are dated

Read the 2016 Service Level Agreement for the IANA Numbering Services cold and the first surprise is how ordinary it is. Allocation requests are acknowledged within two Business Days. A request for missing information follows within four Business Days after acknowledgment. Registry implementation completes within four Business Days after acknowledgment or after the missing information arrives. Performance Standards Metric Reports go out monthly, no later than fifteen calendar days after month-end, alongside an annual service survey and an annual security and systems audit.

Then Amendment No. 1, dated 12 November 2024, added reverse-DNS resolution services and attached numbers to them: API uptime of at least 99% excluding scheduled maintenance, distribution servers available at least 99% of a calendar month measured monthly, and propagation to those servers within ten minutes for 99% of API requests. A miss now triggers a post-mortem shared with the registries, covering the event, the perceived damage, the cause, the preventive measures and a schedule for the next ones. Monthly compliance reporting stays on the same fifteen-day clock.

Two seams run through that amendment and matter more than they look. The five-minute acknowledgment wording is conditioned on where an interface is configured to acknowledge, while the monthly reports express the same metric as no more than 300 seconds for 99% of requests. And DNS-query availability is written as a best-efforts obligation rather than a flat availability floor. The signature page of the copy retained for this review shows a blank AFRINIC line; the amendment's complete execution record was not obtained here.

Who decides that a miss is a failure

The remedy is not decorative. A materially established failure can open a thirty-calendar-day cure period, and if the failure survives the cure, termination becomes available, with successor-transition provisions attached to that ending. The ICANN overview page describes the instrument as carrying performance expectations together with escalation and dispute-resolution mechanisms for non-performance. That is where the consequence lives: in the contract.

The review layer is a different animal. The IANA Numbering Services Review Committee has fifteen members, three appointed by each of the five registries. It meets at least once a year, produces an annual report evaluating performance, and when it finds service-level deficiencies assesses them and reports findings and recommendations to the NRO Executive Council. A separate NRO page on the IANA function puts it the same way: the committee advises and assists the Executive Council. The Number Resource Organization hosts that arrangement. The counterparties named in the executed text are the five registries, and ICANN appears there as the Operator.

So the split is structural, not scandalous. The body that publishes what happened does not hold the lever that converts a miss into a cure. The lever sits with parties who read the committee's findings. A reasonable designer would call that separation of assessment from remedy. A reasonable sceptic would ask who has ever pulled it.

The record prints its own variance

In June 2025 the operator's own monthly performance report printed a row reading, as reported, "DNS Propagation Time | ≤ 10m (99.00%) | 42.84m", against 130 implementations. Availability for the API and for distribution servers both showed 100% against 99% expectations. The summary table does not say whether 42.84 minutes is a mean, an aggregate or another statistic; the detailed table on the same page lists mean, median and maximum separately. Whatever the statistic, the variance was published rather than buried, and that is the part of this design that deserves credit.

The verdict came later. The 2025 IANA Numbering Services Review Committee Report, published 30 March 2026 and covering 1 January to 31 December 2025, worked from the monthly reports and an annual performance matrix. It reports, in its own words, that "the SLA was met 100% of the time in all cases except for reverse DNS performance, where the SLA was met 80% of the time." It identified the June spike in propagation time, concluded that operator performance was acceptable, and found no issue with the ability to deliver the services. An eighty per cent figure and an acceptability conclusion sit a few pages apart, and the report offers no arithmetic that closes the gap. Read as a whole, "the SLA was met" functions as an overall judgement, not a per-metric compliance claim. The separate 2025 performance matrix cited by that report was not read for this review.

A hundred per cent of nothing

Then there is October 2026. That month's performance report records no allocation requests and no reverse-DNS implementations, prints "Requests acknowledged on time (100%)", and shows API and distribution availability at 100% against a 99% expected minimum. The transaction-level categories show dashes instead of percentages. The page states its reporting period but carries no separate publication date.

A percentage printed over zero observations is not a performance result. It is either vacuously true or undefined, and the dashes in the same table show that the reporting system knows the difference between a month with no transactions and a month with measured compliance. Whoever laid out that page chose a dash in one row and 100% in another for the same empty month.

Where the judgement actually sits

Put the three items together and the design becomes legible. Obligations are dated and measurable. Reporting is genuinely public, including the month that missed. The conversion of a variance into a cure or an ending runs through a "materially established failure" determination, and the reviewed record contains no text showing that such a determination was made, declined or weighed after June 2025. The reviewed text does not explain the acceptability reasoning either. The advisory mandate answers the question of who reviews.

It does not answer the question of who enforces, and the reviewed pages give no indication that it was built to.

For readers who rely on this record, the consequence is narrow and practical. A published percentage is evidence of measurement, not of consequence. A dash and a 100% can describe the same uneventful month. A clean annual finding is compatible with a documented metric miss, because the two statements answer different questions.

What would change the reading: a stated threshold that turns a variance into a failure without a discretionary finding; a published record of a cure period opened; a report that prints its denominators beside its percentages. Until one of those appears, the honest description is a compliance ledger with a gated remedy, disclosing more than most service contracts and promising less enforcement than its termination clause suggests.