Summary

  • AFPUB-2026-v6-001-DRAFT03, submitted on 17 September 2026 and still under discussion, changes the proposed IPv6 compliance tests attached to additional IPv4 requests. It does not report an implemented rule or an observed compliance result.
  • Draft 2 measured access networks through selected traffic destinations and hosting through AAAA availability plus IPv6 reachability. Draft 3 instead counts customers with global IPv6 delegations for access, and IPv4 space covered by IPv6 assignments for hosting.
  • Because the new draft keeps percentage-shaped milestones while replacing both denominators, a future pass or failure needs a versioned measurement receipt. “95%” alone is no longer a stable fact.

The same number, a different observation

A result of 95% looks portable. It invites comparison across operators, years and policy revisions. In AFRINIC’s latest IPv6-for-IPv4 proposal, that portability is precisely what has disappeared.

The third draft was submitted on 17 September after mailing-list discussion and a staff impact assessment of the previous text. Its revision history says that the criteria changed. That short note carries more operational weight than the unchanged shape of the milestones suggests.

Under Draft 2, an access provider’s plan began with its actual IPv4 top-25 traffic destinations. For the external destinations that were IPv6-enabled, the proposed thresholds were 25% at 12 months, 50% at 24 months and 75% at 48 months. The AFRINIC-37 minutes record the intended observable as IPv6 traffic volume to those destinations.

Hosting used another construction. Services, applications or content were to be tested through the availability of AAAA records and IPv6 reachability from the Internet. The proposed schedule was 25% at 12 months, 75% at 24 months and 95% at 48 months.

Draft 3 removes both constructions. For an access network, the counted unit is now the customer. The proposal asks what percentage of customers have access to global IPv6 delegations without intervention from the network operator: 25% at 12 months, 50% at 24, 75% at 36 and 95% at 48. Its note connects that phrase to access devices permitting a working IPv6 configuration and the assigned space being reachable from the global routing table.

For content hosting, the denominator is no longer a set of services or DNS records. It is the percentage of assigned IPv4 space that is also covered by IPv6 assignments, on the same four-step timetable. The explanatory note again uses reachability from the global routing table.

These are defensible policy proxies to debate. They are not synonyms. A delegation is not customer use. Address-space coverage is not traffic. A reachable assignment is not proof that an application works end to end. The proposal may choose one of those observations for compliance; a later record must say which observation it chose.

The new schedule adds a 36-month checkpoint and carries 95% into both provider classes. A hybrid provider offering access and hosting would have to satisfy the two tests independently. That operator could therefore have two percentages on the same date, derived from different counted units. Combining them into one score would erase the design of the draft itself.

Draft 3 also names the clock. For an additional IPv4 request, time runs from the first issuance of IPv6 space to the LIR or End User, or from the policy implementation date, whichever is later. This is clearer than leaving the baseline implicit, but it makes the baseline date part of the result. Two operators measured in the same month may be at different policy ages.

The enforcement sentence has changed too. Draft 2 treated failure to comply with the plan as a policy violation. Draft 3 says failure, once verified by AFRINIC, would constitute a violation. It says AFRINIC would establish automated testing as much as possible, supplemented by public tools and measurements. It does not say that the automation already exists, identify its version, or publish a finished evidence contract.

That distinction matters because the staff assessment for Draft 2 raised questions about denominators, validation, acceptable evidence, partial compliance and justified delay. The assessment also identified possible effects on MyAFRINIC, transfers, billing and member-service procedures. It is evidence about Draft 2, not an impact assessment of the changed Draft 3 criteria.

The proposal remains under discussion. The checked record does not show consensus, Last Call, Board ratification, implementation, an applicant measurement, a denied request or a sanction. The present question is therefore not whether any member passed. It is whether the policy record is being designed so that a future result can be interpreted.

A compact measurement receipt would do that work. It should name the proposal or policy version and effective date; the operator class; the clock-origin event; the measurement window and snapshot; the counted-unit definition; denominator and numerator; the delegation, assignment and reachability rule; the tool or method version; the evidence custodian; bounded exclusions; and the current decision state. Corrections should supersede an earlier receipt rather than silently rewriting it.

This is not a request for a new regulator or for disclosure of customer identities. Aggregate counts and reason codes can preserve privacy. The point is narrower: when the denominator changes, the label on the result must change with it. Otherwise the registry may publish an exact percentage that cannot be compared with the percentage beside it.

Sources