Summary

  • AFPUB-2026-v6-001-DRAFT02 is Under Discussion and was returned to the RPD mailing list after AFRINIC-37; it is not an implemented eligibility rule.
  • The draft names 12-, 24- and 48-month IPv6 percentages, but the public text and staff assessment do not yet establish a reproducible measurement record.
  • External-destination traffic and hosted-service reachability are different measurement branches, with different denominators and failure modes.
  • A privacy-protecting receipt could make a narrow decision auditable without publishing raw flow records, customer identities or network topology.

The strongest benign case is that the draft tries to connect scarce IPv4 access with a credible transition plan. For a requester without IPv6 space, the proposed text would require an IPv4 request to be accompanied by an IPv6 request and a coherent IPv6 deployment and addressing plan. That is not an instruction to redesign an operator’s network. It is a proposed condition in AFRINIC’s private number-resource policy process, not legislation, a court order, a route change or a national IPv6 mandate.

For external destinations, the draft refers to the requester’s actual top 25 IPv4 traffic destinations that are IPv6-enabled. It sets minimum IPv6 percentages of 25% within 12 months, 50% within 24 months and 75% within 48 months. For hosted services, applications or content, it uses a separate branch: the share of AAAA records that are available and reachable over IPv6, with targets of 25%, 75% and 95% at the same intervals. Those branches should not be collapsed into one score. Outbound traffic to external destinations measures one relationship; the reachability of services controlled or hosted by the requester measures another.

The gap is methodological. AFRINIC staff said the mechanism for identifying and measuring the top 25, validating IPv6 enablement and assessing the percentages is not specified. The draft also does not define acceptable evidence for compliance, partial compliance or justified delay. At AFRINIC-37, the author described NetFlow or similar tools as a simple way to observe traffic to major destinations and suggested that a network with fewer than 25 destinations could use the available set. That was an attributed proposal, not proof that NetFlow is mandatory, universally deployed, unbiased or sufficient.

A percentage is only meaningful with its universe. The result can change according to whether the observation includes transit, peering, access, cloud egress, private addresses, sampled records or only selected interfaces. A busy week, a seasonal month and a twelve-month window may produce different top destinations. The selection rule also matters: rank by bytes, flows, sessions, customers or distinct destinations, and ties can produce different lists. If the top 25 are selected after IPv6 capability is known, the procedure can reward a different choice than if destinations are selected first and tested afterward.

“IPv6-enabled” needs an operational test. It might mean an AAAA record exists, that a route is visible, that a connection succeeds from a defined vantage point, or that observed traffic uses IPv6 during a stated interval. Each test answers a different question. Hosted services add further choices: which records count, how aliases and CDNs are treated, whether temporary failures are retried, and whether the denominator is records, services, names or reachable endpoints. Weighting can likewise privilege a large service, a small customer or an individual record.

Exclusions must be visible. Maintenance, outages, abuse controls, encrypted telemetry, missing collectors, private destinations and data-protection restrictions may all affect the denominator. Tool configuration, sampling rates, collector clocks, software versions and retention policies can change an apparent result. Without those facts, two operators could report the same percentage from different universes and still be unable to explain the difference later. That is a bounded inference from the missing specifications, not a finding that anyone has manipulated a metric or caused harm.

A workable compromise is a measurement receipt, not publication of raw flows. The receipt could identify the requester and holder to the authorized reviewer, record the draft or policy version, measurement class, observation start and end, eligible universe, top-destination rule and count, IPv6-enabled or reachable test, weighting or hosted-record denominator, exclusions and reason codes, tool and material configuration versions, applicable threshold, observed percentage and a calculation fingerprint.

It should also name the evidence custodian and reviewer, record any partial-compliance or external-delay explanation, and preserve the decision, reasons, review path and correction history.

The public part could disclose definitions, dates, denominator class, hashes, result ranges, reasons and corrections. Raw flow records, customer identities, risky destination details, device configuration and topology should remain protected. A receipt cannot solve every dispute, but it can allow a later reviewer to replay the decision inputs without exposing operational secrets.

Verified process facts must remain separate from advocacy. AFRINIC lists Draft 2 as Under Discussion; it was submitted on 14 June 2026 and would amend the Consolidated Policy Manual. The AFRINIC-37 record says the proposal did not reach rough consensus and was sent back to the mailing list. There has been no claimed implementation, eligibility decision, denied request, breach finding or sanction under this draft. Regional capability figures in the proposal are not this article’s fresh measurement.

Sources

https://www.afrinic.net/afpub-2026-v6-001-draft02.html https://www.afrinic.net/policy-proposals.html https://www.afrinic.net/ppm-afrinic-37.html