Summary

  • The five RIRs can demand cooperation with reviews, issue a dispute notice, enter mediation, commence binding arbitration, seek temporary judicial relief, decline renewal, terminate after an uncured arbitral finding of material failure, and select a successor operator. Those powers belong to the RIRs acting collectively and unanimously, not to the advisory Review Committee or to one registry acting alone.
  • Monthly reports since March 2017 have made request timing, accuracy and reverse-DNS performance visible. The record through 2024 was largely uneventful and low-volume. The 2025 review, however, reported reverse-DNS performance at 80%, including a June propagation result of 42.84 minutes against a ten-minute expectation, while still concluding that the operator's overall performance was acceptable.
  • A metric miss is not self-executing. Article 9 requires discussion, a judgment that failure is material, mediation and then arbitration before the thirty-day cure period linked to termination begins. The agreement contains no schedule of automatic service credits, published severity classes or mandatory public correction notices for each missed target.
  • The central weakness is therefore not the absence of ultimate rights but the distance between observation and cure. A stronger regime would preserve operational discretion while publishing incident explanations, severity findings, correction deadlines, repeated-miss rules and successor-readiness evidence before a problem becomes serious enough to justify replacement.

The contract that replaced a distant steward

When United States stewardship ended on 30 September 2016, the numbering community did not accept a vague assurance that the same technical staff would continue doing good work. AFRINIC, APNIC, ARIN, LACNIC and the RIPE NCC had signed a service-level agreement with ICANN on 29 June. It became effective on 1 October, once ICANN was released from the relevant obligations in the United States government's IANA functions contract and accepted responsibility for continuing the numbering services.

The design reflected a clear institutional judgment. The direct customers of the top-level numbering service were the RIRs. They request allocations of IPv4, IPv6 and autonomous system number blocks under global policies; they depend on accurate top-level registries; and they rely on the associated reverse-DNS zones. If stewardship was to move away from a national government, standing should move toward those customers rather than dissolve into the broad reputation of ICANN.

That move was more consequential than changing the name on an oversight chart. The agreement named parties, duties, evidence, time limits, dispute procedures, governing law and continuity obligations. It required the RIRs to reimburse ICANN for the cost of the service, initially fixed at $650,000 a year unless all parties agreed otherwise. It required five-year terms with automatic renewal, allowed non-renewal and provided a path to a successor. The RIRs became legal counterparties instead of commentators asking another steward to intervene.

The bargain also preserved a separation that is easy to lose in public debate. ICANN was the operator, not the author of number-resource policy. The agreement describes the service as administrative and technical. It requires stable and secure performance according to global policies developed through the numbering community's established mechanisms. Service staff may explain operational questions and ask for clarification, but they are not supposed to campaign for a policy outcome. Accountability therefore has two targets: faithful execution by the operator and legitimate policy development by the affected community.

The agreement addresses the first without confiscating the second.

This is why the document deserves to be read as a constitution for a narrow service, not as a general statement of Internet values. It gives the RIRs real power, but only over specified numbering obligations. It also reveals the limits of contractual accountability when the service is highly reliable, the transaction count is small and the strongest remedy could itself endanger continuity.

What the numbering operator actually promises

The covered service sits at the top of the address and autonomous-system-number hierarchy. IANA maintains the unallocated portions of the global unicast IPv4, IPv6 and ASN registries, records allocations and returns, and supports the unicast portions of IN-ADDR.ARPA and IP6.ARPA. The RIRs then administer resources for their respective regions under regional and global policies. The operator does not assign every address to every network, and an RIR does not need ICANN's permission for each regional member decision.

For a valid RIR request, the original agreement established a sequence of practical obligations. ICANN must acknowledge the initial request within two business days. If more information is needed, it must make a specific and complete request within four business days of acknowledgement. If the request will be fulfilled, the registry should be updated within four business days after acknowledgement or receipt of the additional information. The requesting RIR and the other RIRs are then informed, and the public records are updated.

If ICANN is unable or unwilling to fulfill a request within ten business days, it must explain the status and reason in writing and continue providing updates. A valid request that remains unsatisfied after thirty business days may be asserted as a failure to perform if Article 9's conditions are met. The wording matters. Thirty days does not automatically terminate the operator or even automatically establish material breach. It creates an evidential threshold from which the collective RIRs may begin the formal path.

Accuracy is as important as speed. A right-sized allocation must not overlap an existing allocation and must match the intended delegation. The top-level registry must be updated correctly before an RIR relies on the result. A quick but wrong allocation would create a uniqueness problem across the global system; a slow but correct allocation could impair planning without corrupting title. A useful service regime must see both kinds of harm rather than treating elapsed time as the whole service.

The operator must also keep relevant public registry information in the public domain and provide the RIRs, on request, with registry data, open-request information, request history and other non-proprietary material necessary to continue the service. These obligations are not mere transparency accessories. They reduce the incumbent's ability to make itself irreplaceable by holding the authoritative record or operating history in an unusable form.

Security duties complete the promise. The operator must use secure communications with the RIRs, preserve the authentication, integrity and reliability of data, maintain a security plan, notify the RIRs of outages and make annual security-audit material available. The resulting duty is broader than hitting a response-time target. A service could answer every request on time while exposing credentials or weakening the integrity of a registry. Such conduct could still amount to material non-performance even if the monthly timing columns remained green.

Measurement arrived after commencement

The agreement required ICANN and the RIRs to develop public informational performance reports and publish them monthly, no later than fifteen calendar days after each month. It allowed up to six months after commencement for that reporting arrangement to begin. The IANA archive consequently starts its post-transition number-resource series in March 2017 rather than October 2016.

That delay was contemplated, but it creates the first boundary in the historical record. The agreement was effective from October; the comparable monthly matrices began later. The first Review Committee report treated performance before March 2017 as outside its usable measurement period because the agreed indicators had not yet been developed. Assertions about uninterrupted numerical compliance from the first day of the transition therefore go beyond the evidence. The service continued, but the modern reporting series did not yet exist.

Once established, the monthly reports gave the community a compact view of allocation requests. They recorded acknowledgement, response, implementation and accuracy. Later reporting also covered the upper reverse-DNS service: acknowledgement of automated change requests, propagation time, service availability and authoritative DNS availability. The archive is unusually accessible for a quiet coordination function. A reader can move month by month from 2017 to the present and see whether requests occurred and how the operator reported them.

The value of this record should not be understated. A service-level promise without a dated record can be rewritten after the fact. Monthly publication fixes a contemporaneous observation. It lets the RIRs compare their own correspondence with the operator's account. It gives the Review Committee common evidence. It also makes long periods with no allocation requests visible, preventing a perfect percentage from being mistaken for a large sample of difficult work.

But publication does not settle interpretation. A percentage can describe only the denominator it was built to count. No allocation request in a month may yield no timing result, not proof that the operator succeeded at an allocation. A request may be technically completed while a customer remains dissatisfied with communication. An availability figure may remain perfect while one propagation episode exceeds its target. The contract therefore needs human review, yet human review reintroduces discretion between the observed number and the legal consequence.

A mostly quiet record from 2017 through 2024

The first annual review covered April through December 2017. It found two IPv4 requests and two ASN requests, all completed accurately and on time. The committee reported no failure or near failure and said it had received no comments during its public consultation. That combination established a pattern that would recur: few top-level allocations, successful handling, an annual call for input and little or no public response.

The low volume is structurally unsurprising. The RIRs do not return to IANA for each allocation to a network. They request much larger top-level blocks when a global policy and their inventory position require it. IPv4 exhaustion further reduced ordinary fresh allocations. The absence of a busy transaction stream is therefore evidence about the architecture, not neglect by itself.

Annual reports from 2018 onward repeatedly concluded that the service was acceptable and that no significant concern required escalation. In 2023, for example, the committee recorded only two ASN allocations, one requested by APNIC and one by the RIPE NCC, and said the measured service levels were met in all cases. In 2024 it recorded two ASN requests, from LACNIC and ARIN, and one IPv6 request from APNIC, again concluding that all measured obligations were satisfied.

This history supports a strong but bounded conclusion: the public record shows sustained reliable operation of the allocation service. It does not test every remedy. There is no visible trail of a formal Article 9 dispute, mediation, arbitral finding, thirty-day cure or operator termination in the reviewed official materials. Good performance is one explanation. Another is that ordinary concerns can be solved informally before the legal sequence begins. The archive does not reveal every private conversation, and it should not be used to invent one.

The public-comment pattern also requires care. Repeated calls often received no comments. That may indicate satisfaction, low salience, confidence in RIR staff review or simple attention scarcity. Silence is not a vote with a known electorate. It cannot prove that every network operator in every region independently examined the matrices and approved the result. The committee's regional composition supplies an important representative check, but the direct operational evidence still comes heavily from RIR staff and the operator's own reports.

Until the reverse-DNS service was added to the explicit framework, the observed allocation series also tested a relatively sparse activity. A regime can look flawless when the measured event count is two or three a year. That does not make the result false. It means conclusions about resilience under frequent operations need evidence from services with a richer event stream.

The 2024 amendment changed the evidence base

On 12 November 2024, the parties amended the agreement to incorporate reverse-resolution services. This was a significant expansion of what could be measured under the same accountability relationship. Reverse DNS translates an address into a domain-name context through in-addr.arpa for IPv4 and ip6.arpa for IPv6. The RIRs submit upper-level changes associated with their number holdings, and IANA propagates them into the relevant zones.

Unlike occasional top-level allocations, reverse-DNS changes can occur many times in a month. The June 2025 report counted 130 implementations. The revised indicators included acknowledgement within 300 seconds for 99% of requests, propagation within ten minutes for 99%, availability of the request interface, availability of the distribution system and availability of authoritative DNS service.

The amendment was a governance improvement because it brought a dependency already connected to numbering into explicit measurement, continuity and review. It also exposed a methodological problem that a low-volume allocation table had rarely forced the committee to confront: what should happen when one metric clearly misses but the service remains available and users suffer no demonstrated broad outage?

June 2025 provides the clearest public case. IANA reported no allocation requests that month. For reverse DNS, the service and distribution systems were fully available, as were the listed authoritative DNS endpoints. Yet reported propagation performance was 42.84 minutes against an expectation of ten minutes for 99% of changes. The same report showed a mean of 20.10 minutes across 130 implementations, a median of 4.92 minutes and a longest time of 44.20 minutes.

Those figures support several observations at once. Most users could still resolve reverse DNS; availability was not lost. The median being below ten minutes indicates that many changes propagated promptly. Yet the high mean, long maximum and 42.84-minute reported performance show a substantial tail and an unambiguous miss against the stated threshold. A green availability result cannot erase a red propagation result because the two measure different promises.

The 2025 annual review, published in March 2026, did not hide the issue. It said RIR staff confirmed 100% performance in all cases except reverse DNS, where the target was met 80% of the time, and specifically noted the June spike. It also reported that no community comments were received. The committee nevertheless concluded that performance under the agreement was acceptable, that the SLA was met and that no issue was observed concerning the operator's ability to provide the service.

This is the enforcement gap in its most useful form. The number was published. The annual reviewer repeated it. The conclusion did not describe a formal breach, a required correction deadline or a continuing corrective measure. Nothing in the published result shows that the committee acted irrationally; a transient propagation delay may not be material failure. But the reader cannot see the rule that converted an 80% annual success rate into an acceptable overall judgment.

A miss is not the same as material failure

Service-level language often creates the impression of automaticity. A target says ten minutes; an observation says forty-two; therefore breach and remedy follow. The 2016 agreement is more conditional. Article 9 is triggered when the RIRs believe the operator has materially failed to perform required services or materially breached the agreement. Materiality is a legal and operational judgment, not a percentage cell.

That distinction is defensible. Not every late response threatens the Internet. A single measurement anomaly, a dependency outside the operator's reasonable control or a delay with no lasting harm may deserve explanation and correction rather than litigation. Automatic termination for one missed minute would make the service less stable, not more accountable. The word "materially" protects proportionality.

Yet proportionality needs an intermediate vocabulary. The agreement does not publish a severity ladder distinguishing an isolated miss, a repeated miss, a near failure, a customer-impacting incident and a material failure. It does not say that two consecutive propagation failures require a public correction plan, or that a repeated accuracy problem automatically advances to independent review. It leaves the movement from metric to dispute largely to collective judgment.

The 2025 report illustrates the cost. A reader can see that a target was missed and that the result was still acceptable, but not the evaluative bridge. Was June caused by a planned change? Did the operator correct a configuration? Did no RIR request suffer practical harm? Was 80% calculated across five measured months after the amendment rather than a full year? Which condition would have changed the conclusion? The selected public documents do not answer all of these questions.

This is not a demand to publish sensitive security details. It is a demand for governance facts: classification, duration, affected service, customer effect, correction status and the reason formal escalation was or was not warranted. Those facts can often be disclosed without revealing credentials, defensive design or confidential correspondence.

The current regime is consequently strong at observability and weak at deterministic consequence. It can prove that a target was missed. It does not automatically prove a material failure, identify a cure or impose a visible sanction. That gap is the space in which expert judgment operates, but also where accountability can become too dependent on trust among institutional peers.

The Review Committee can warn but cannot compel

The IANA Numbering Services Review Committee was created to advise and assist the NRO Executive Council. Its charter requires it to report any concern, including observed failure or near failure, at least annually. It has fifteen seats: two community appointees from each RIR region and one non-voting RIR staff member from each region. Its proceedings are generally open, its reports are exposed for comment, and its regional composition prevents one registry from supplying the whole judgment.

These are valuable safeguards. Community appointees can ask whether a technically correct service still meets user expectations. Staff representatives can compare the public numbers with actual requests. Equal regional representation prevents transaction volume from giving the busiest region all the voice. The annual report creates a recurring moment at which apparently routine performance can be questioned.

But the committee is expressly advisory. Its charter says it cannot commit the NRO EC to a recommendation or decision unless authority has been specifically delegated. It is not a contracting party. It cannot itself serve a binding dispute notice on ICANN, commence arbitration, terminate the agreement or appoint a successor. Those rights belong to the RIRs under the agreement.

This separation is institutionally reasonable. Community reviewers should not be able to end a vital service casually. The RIRs carry operational responsibility and financial obligations, and their executives must plan continuity. Yet the separation also creates a possible choke point. A committee may identify a near failure, but a remedy depends on the NRO EC and all five RIRs translating advice into collective legal action.

The quality of accountability therefore depends not only on what the committee publishes but on how the NRO EC responds. A mature public record would show that a concern was accepted, rejected or sent for correction, with reasons and a date for follow-up. In years of full compliance, a short acceptance is enough. In a year with a visible miss, silence about the evaluative bridge makes the advisory design look thinner than it may be in practice.

The committee's open consultation is also not a substitute for power. Public comments can reveal harm and challenge a complacent conclusion, but repeated zero-comment periods make the mechanism a weak detector on its own. Users who rely on the RIRs may not monitor a specialist annual matrix. The strongest evidence will usually come from direct customer records, synthetic measurements and incident notices, not from waiting for an outsider to email a concern.

Unanimity protects regions and slows action

Article 3 contains one of the most consequential clauses in the agreement: the RIRs may exercise their rights and powers only collectively and unanimously. Although rights benefit them jointly and severally in other respects, the exercise of contractual power requires all five to act together.

The legitimacy case for unanimity is strong. The service is global. No single RIR should be able to terminate the common operator and impose transition risk on four other regions. AFRINIC's view cannot be treated as expendable because its recent allocation count is lower. LACNIC or APNIC should not discover that another registry changed the top-level service without their consent. Unanimity gives each region protection against unilateral disruption.

The enforcement cost is equally real. One RIR may experience harm that others do not. A delay could coincide with a time-sensitive policy implementation in one region. A governance crisis could impair one registry's capacity to approve action. A disagreement over materiality could hold the collective below the threshold for a dispute notice. Even when every party wants correction, negotiating a common legal position consumes time.

This design makes informal correction especially attractive. If operator and RIR staff can resolve a problem through technical discussion, there is no need to assemble five unanimous approvals for adversarial steps. That is efficient during ordinary operations. It also means the public enforcement record may remain empty even when meaningful corrections occur.

The danger appears when consensus itself becomes the remedy gate. A global operator can be held to account only if all direct customers agree not merely that a target was missed but that the failure is material and worth escalating. The agreement does not provide a lower-threshold public corrective measure that two or three RIRs can invoke while leaving termination unanimous. It therefore bundles mild and severe action too closely around collective decision-making.

A more graduated arrangement could preserve unanimity for non-renewal, termination and successor selection while allowing a qualified subset to require an incident statement, an independent measurement review or a time-bound correction plan. That would not let one region destabilize the service. It would make smaller remedies easier to use before a disagreement hardens into a global crisis.

The formal remedy ladder is long by design

If informal resolution fails, the agreement provides a detailed ladder. First, an aggrieved party sets out the dispute in writing. It should use reasonable efforts to do so within thirty calendar days after learning of the cause. If no notice is delivered within one year, rights relating to that dispute are deemed waived. The one-year limit prevents indefinite resurrection of an old operational event, but it also rewards timely recognition and documentation.

After receipt of the notice, the parties attempt a mutually agreeable resolution. If they cannot resolve the matter within sixty days after their first meeting, the dispute moves to mediation. The mediator must have an appropriate legal or judicial background, familiarity with California contract law and experience with the Internet sector. The mediation is confidential and treated as a settlement discussion.

If mediation has not resolved the dispute sixty days after selection of the mediator, either party may elect arbitration under the International Chamber of Commerce rules. Three arbitrators are selected, each residing in a different RIR region. They must combine legal or judicial qualification with relevant technical and governance experience. Proceedings are to take place in Geneva unless the parties agree otherwise, and the parties are to use commercially reasonable efforts to obtain an award within 120 days of the designated procedural starting point.

An arbitral majority's decision is final and binding. A party can go to court to enforce the award. Temporary or preliminary injunctive relief is also available before or during arbitration to preserve rights. These provisions answer the criticism that the agreement is merely aspirational. It creates an adjudicative path outside ICANN's own board and outside any single RIR.

Termination, however, requires more. The arbitrators must determine that ICANN materially failed to perform or materially breached the agreement. ICANN then receives thirty calendar days to remedy the failure. Only if the RIRs conclude that cure was inadequate may they terminate immediately, and their exercise of power remains collective and unanimous.

This sequence values continuity over speed. Discussion, mediation and arbitration create multiple chances to correct an issue before replacement. For a policy disagreement dressed up as a service complaint, independent arbitrators can keep the RIRs within the contract. For a genuine but reparable incident, the operator can restore performance without exposing the global system to an abrupt handover.

The same sequence is poorly suited to an urgent operational failure unless interim relief and voluntary cooperation do most of the work. Thirty days for notice, sixty days of discussion, mediator selection, sixty days of mediation, arbitration and a thirty-day cure can extend over many months. Some periods are goals rather than automatic deadlines. A vital service cannot remain broken while the calendar runs. The legal ladder is therefore an ultimate accountability device, not an incident-response substitute.

Non-renewal is easier in law and difficult in practice

The agreement runs for five years and renews automatically for further five-year terms. Either side may elect not to renew without first proving material failure, provided it gives notice at least twelve months before the current term ends. This is the cleanest exit right because it avoids litigating breach.

Non-renewal still requires planning. The RIRs must identify and agree on a successor. That operator needs secure systems, trusted staff, connectivity, documented procedures, registry data, communication channels and acceptance across all five regions. The incumbent must cooperate with a detailed transition plan and use best efforts for an orderly handover. The RIRs reimburse approved transition expenses.

The agreement anticipated this problem by requiring a transition framework within 180 days after commencement. Once non-renewal is notified or arbitration begins, the RIRs may notify ICANN of their selected successor. ICANN must then cooperate on a detailed plan, generally due within ninety days after selection. After termination, it must continue best-efforts transition assistance for up to ninety days.

Portability provisions strengthen this remedy. Public registry data remains public. Essential records, metadata, correspondence and history must be maintained in non-proprietary form. The operator should not be able to claim exclusive ownership over the facts required to continue the numbering service. Contractual portability turns replacement from an abstract threat into a conceivable operation.

Yet conceivability is not readiness. The public record reviewed for this article does not identify a standing alternative operator that has repeatedly demonstrated it can assume the service on short notice. Data can be portable while expertise, trust and secure operating arrangements remain concentrated. PTI performs the work under arrangements within the ICANN corporate family, and other IANA services share people and infrastructure efficiencies. Separating one function may require careful division of systems without weakening the rest.

Replacement is therefore both the strongest remedy and the remedy most likely to be withheld until failure is severe. This is a familiar infrastructure paradox. The more essential and specialized the incumbent, the more dangerous termination becomes; the more dangerous termination becomes, the less credible it is as discipline for ordinary misses. Frequent, low-cost corrective remedies are needed precisely because the ultimate remedy is so expensive.

ICANN is the counterparty, PTI performs the work

The 2016 SLA names ICANN as the operator. After the transition, Public Technical Identifiers, an ICANN affiliate, performs the IANA functions under separate arrangements. A subcontract passes the numbering work from ICANN to PTI, but it does not replace the RIRs' principal counterparty.

This structure preserves a clear pocket of responsibility. ICANN cannot answer an RIR complaint merely by saying that an affiliate made the operational mistake. It promised the service. The RIRs can pursue their rights against ICANN, and ICANN must manage its relationship with PTI. Corporate separation supplies focused governance and accounts for the operating entity without forcing the five RIRs to renegotiate their core standing.

It also adds distance between measurement and control. The Review Committee observes the service delivered by IANA staff under PTI. The contract right is against ICANN. Corrective action may involve PTI management, ICANN management, shared services and the RIRs. In routine cases this distinction may be invisible. In a severe dispute, the legal and operational lines matter.

The arrangement differs from the naming function, where ICANN holds a dedicated naming contract with PTI and the ICANN Bylaws create a Customer Standing Committee, naming-function reviews and a naming-specific separation procedure. Numbering customers should not assume those mechanisms enforce their SLA. Their rights arise from the ICANN-RIR agreement, their Review Committee charter and their own collective action.

This distinction also matters for replacement. Terminating the SLA ends ICANN's authority to provide the numbering service under that agreement; it does not automatically dissolve PTI or transfer the naming and protocol-parameter functions. The RIRs select a successor for numbering. Other operational communities retain their own agreements and decisions. A remedy can therefore be function-specific even when the current operator shares staff and facilities.

Incident transparency stops short of the whole cure record

The agreement includes meaningful disclosure obligations. ICANN must publish monthly performance reports, collaborate on an annual service survey, provide annual security-audit material to the RIRs and notify them of outages. The Review Committee publishes an annual matrix, invites comments and issues its conclusion. Meeting archives and committee proceedings are generally open.

Several limits are also explicit or practical. Individual RIR requests begin as confidential communications and cannot simply be exposed to other parties without permission, subject to necessary exceptions. Security plans and detailed audit records are provided to the RIRs rather than automatically published in full. Mediation is confidential. Arbitration may produce an award, but the agreement does not require every preliminary submission or settlement offer to become public.

These limits protect real interests. An allocation request can reveal planning before an RIR is ready to announce it. Security evidence may expose weaknesses. Confidential mediation makes compromise possible. Transparency that damages the service is not accountability.

The missing element is a public, non-sensitive cure record. A monthly page can show that propagation took too long, but it does not necessarily say why, what was changed, whether recurrence was tested or when the matter was closed. An annual conclusion can call performance acceptable without publishing a severity judgment. The public then sees the measurement and the verdict but not the correction.

A compact incident notice could close much of this gap. It could identify the affected commitment, start and end times, scale, user effect, broad causal class, immediate mitigation, long-term correction, validation method and Review Committee disposition. Security-sensitive fields could be withheld with a reason. If no correction was required, the notice could explain why the miss did not reflect operator failure.

Such a notice would not convert the committee into a court. It would let the community distinguish an isolated measurement artifact from a repeated control weakness. It would also protect the operator from exaggerated claims, because a transparent explanation can show that availability was preserved and the issue was contained.

Service credits would not solve the hardest problem

Commercial service agreements often attach automatic credits to missed targets. The IANA numbering agreement does not publish a table in which a late acknowledgement produces a fixed payment or a propagation miss reduces the annual reimbursement. That absence contributes to the distance between measurement and consequence.

A credit could create a modest automatic response without invoking material breach. It would also be easy to administer. Yet money is not the central risk here. The annual reimbursement is a cost arrangement, not a conventional profit contract. A small credit may not change behavior, while a large one could drain resources needed to repair the service. The real customer interest is accurate, secure and continuous coordination.

The better intermediate remedies are informational and operational. A first miss should produce classification and explanation. A repeated miss should require a correction plan with dates. A serious or accuracy-related incident should trigger independent validation. A persistent pattern should initiate a successor-readiness review even before formal arbitration. Financial consequences, if used, should reinforce these duties rather than substitute for them.

Accuracy deserves a lower tolerance than ordinary latency. A wrong allocation can create overlapping claims and contaminate records across regions. Security and integrity failures may justify rapid escalation even if they are brief. Propagation delay with full availability may warrant a less severe response, but repeated failure should still become material at some defined point. One uniform percentage cannot capture these different harms.

The agreement already gives the parties enough authority to adopt many of these practices by mutual consent. The hard question is whether they will bind themselves before a crisis. Institutional peers often prefer discretion because it supports cooperation. Customers need enough predictability to know that cooperation will not become mutual indulgence.

The 2025 miss is a test of explanation, not a case for removal

The June 2025 result should not be inflated into proof that PTI is unfit. The same monthly report showed full availability, and the annual review found no broader inability to deliver the service. The metric was newly incorporated, the sample covered many changes, and the published record does not establish lasting user harm. Replacement would have been wildly disproportionate on the evidence available.

Nor should the result be minimized. A ten-minute, 99% target is either a meaningful commitment or decorative text. A result of 42.84 minutes, a mean above twenty minutes and an annual statement that the target was met only 80% of the time deserve a public explanation. The fact that the service remained reachable tells readers why the incident may have been non-material; it does not make the propagation promise disappear.

The committee's acceptable-performance conclusion may be correct. Its weakness is that the conclusion carries more reasoning than the published report exposes. The next report should make the standard clear: whether isolated tails are excluded, how the annual 80% denominator was constructed, whether correction occurred, and what repeat pattern would trigger a concern or near-failure finding.

This is how a mature service regime learns from a miss without manufacturing a crisis. It treats the event as a calibration opportunity. It asks whether the target describes customer need, whether measurement is reliable, whether the operator controls the relevant step and whether the consequence ladder is proportionate. It then records the answer so the next reviewer is not starting from zero.

The alternative is a brittle binary. Either every miss is called breach, making the system litigious, or every miss short of catastrophe is called acceptable, making the metric ceremonial. The agreement needs a middle category in which a miss is acknowledged, corrected and tracked without prejudging material failure.

What the RIRs can actually do

The direct answer to the enforcement question is more substantial than critics sometimes suggest. The RIRs can inspect performance through monthly reports and their own request records. They can conduct periodic reviews and require ICANN to cooperate. They can ask for essential data and audit evidence. They can demand outage notification and status updates for delayed requests. They can decide not to renew for any reason with timely notice.

For a dispute, they can deliver written notice, require discussion, proceed to mediation, elect binding arbitration, ask a court for temporary relief and enforce an award. After an arbitral finding of material non-performance, they can assess the thirty-day cure and terminate if it remains inadequate. They can select a successor and require transition cooperation, records and best efforts to preserve continuity.

But every verb has a qualification. Formal RIR power is collective and unanimous. The Review Committee can advise but cannot compel. A missed metric does not automatically equal material failure. Mediation is confidential. Arbitration takes time and money. Termination follows an arbitral finding and cure opportunity. Replacement depends on an alternative operator that can preserve security and continuity.

What the RIRs cannot do under this text is equally important. One registry cannot unilaterally terminate the global arrangement. The committee cannot order ICANN to pay an automatic credit or publish a cure notice that the agreement does not require. A names-community accountability body cannot exercise the RIRs' contractual rights. The existence of a target does not eliminate materiality analysis. Public disappointment does not itself select a successor.

The agreement is thus enforceable, but its practical enforcement is concentrated at the high end. It is strongest when all five RIRs agree that a serious failure justifies external adjudication or planned exit. It is less determinate when a measurable miss deserves correction but not war.

A more credible bridge from evidence to cure

Five reforms would make the existing arrangement more credible without destabilizing it. First, the RIRs and Review Committee should publish a severity framework. Accuracy, integrity, security, availability, timeliness and reporting failures have different consequences. The framework should say what counts as an observation, concern, near failure and presumptive material failure.

Second, each missed target should receive a public disposition. The minimum response can be short: confirmed or measurement error; customer effect; corrected or open; repeat or isolated; escalation not required, monitoring required or formal notice recommended. This would reveal the evaluative bridge while protecting sensitive detail.

Third, repeated misses should create automatic procedural duties even when they do not create automatic legal breach. Two consecutive failures might require a correction plan. A rolling threshold might require independent validation. A recurrence after declared closure might require the NRO EC to explain whether Article 9 is being considered.

Fourth, successor readiness should be tested rather than merely documented. The parties need not run a dangerous full handover. They can verify export formats, contact trees, authentication transfer, zone and registry reconstruction, key dependencies and the ability of a designated contingency team to read the transition material. A remedy is credible only if it can be exercised.

Fifth, the NRO EC should publish responses to Review Committee concerns. Acceptance, rejection and requests for further evidence should be dated. Unanimity should remain for termination and successor choice, while a lower threshold could require explanation or independent measurement. That would protect every region from unilateral removal without allowing one disagreement to suppress mild correction.

None of these reforms requires a presumption of bad faith. They are what allows good-faith cooperation to remain legitimate when personnel change and memories fade. An agreement should not depend on today's parties always sharing the same informal understanding.

The difference between having a remedy and using one

The post-transition numbering arrangement solved a real standing problem. The parties most directly dependent on top-level number coordination obtained contractual rights. The operator accepted measurable duties, external dispute resolution, cure, termination, data portability and successor assistance. That is firmer than a promise to listen.

Its public operation since 2017 also deserves credit. Reports are regular. Annual reviews span all five regions. Top-level allocation requests have been completed accurately and promptly in the periods examined. The 2025 reverse-DNS miss was visible rather than concealed. A governance analysis that ignores this performance would mistake a design gap for an operational collapse.

The gap is nevertheless consequential. Measurement answers what happened to a defined indicator. Enforcement asks whether the event was material, who can demand correction, by when, under what evidence and with what consequence. The agreement answers the final, catastrophic end of that chain better than the ordinary middle.

The RIRs can ultimately replace ICANN as numbering operator. They cannot do so lightly, separately or quickly, and they should not. Their more urgent task is to make smaller remedies visible and predictable. If the community can see why a miss was accepted, how it was corrected and what recurrence would change the judgment, the SLA becomes a living accountability instrument. If the record shows only percentages and an annual verdict, enforcement remains legally real but publicly remote.

The 2025 result is therefore not an indictment. It is a warning that a reliable service can outgrow a binary review vocabulary. The next decade of numbering accountability should be judged not by whether the parties ever reach termination, but by whether they can demonstrate cure long before termination becomes thinkable.

Sources

  1. Service Level Agreement for the IANA Numbering Services, 29 June 2016 — parties, operational duties, reporting, review, dispute resolution, cure, termination, continuity and governing law.
  2. ICANN overview of the numbering SLA — transition purpose, effective date, renewal and successor summary.
  3. NRO numbering SLA page and 2024 amendment — current agreement record and incorporation of reverse-resolution services.
  4. Internet Number Community transition proposal — rationale for direct RIR standing and the Review Committee.
  5. IANA Numbering Services Review Committee charter — advisory authority, composition and annual reporting duties.
  6. IANA Review Committee operating procedures, version 2.1 — membership, governance and review procedures.
  7. IANA number-resource monthly performance archive — monthly reports from March 2017 onward.
  8. 2017 IANA Numbering Services Review Committee report — first comparable review period, request count and findings.
  9. 2022 IANA Numbering Services Review Committee report — mature annual-review method and acceptable-performance finding.
  10. 2023 IANA Numbering Services Review Committee report — two ASN requests, full reported compliance and no public comments.
  11. 2024 IANA Numbering Services Review Committee report — allocation volume and full reported compliance in 2024.
  12. IANA Number Resource Performance Report for June 2025 — reverse-DNS propagation result, implementation count and availability measures.
  13. 2025 IANA Numbering Services Review Committee report — 80% reverse-DNS performance, June spike, no comments and acceptable overall conclusion.
  14. RFC 7020, The Internet Numbers Registry System — allocation hierarchy, IANA role, RIR responsibilities and community-developed policy structure.
  15. PTI agreements index — relationship among the numbering SLA, ICANN-PTI subcontract and other IANA function agreements.