Summary

  • AFRINIC's public 2026 closure data contain one aggregate row labelled VOLUNTART in each of the ASN, IPv4 and IPv6 files, while the established label VOLUNTARY remains present.
  • The downloadable CSV files and the live JSON files agree exactly, and the dashboard groups the literal status string. The extra category is therefore in the published data model, not merely painted onto one chart.
  • The evidence says nothing about whether an underlying return or closure was valid. AFRINIC can correct the public classification without erasing history by publishing a stable status code, an authorised mapping and linked old-and-new dataset hashes.

Two words sit in AFRINIC's current closure statistics. One is VOLUNTARY. The other is VOLUNTART.

The difference is one final character. A reader can guess what probably happened. A statistical system cannot safely guess. In the public files frozen on 13 September 2026, the second token appears once in every resource class: one ASN aggregate, one IPv4 /22 aggregate and one IPv6 /32 aggregate, all dated 2026 and all carrying a total of one. The correctly spelled token also remains in the series.

That is not evidence of a wrongful closure. It is evidence of a public classification fork.

Publication is the first control

The strongest defence of AFRINIC's record comes first. The registry publishes both a downloadable archive and the data used by its web charts. That choice allows an outside reader to reproduce the totals, inspect categories and preserve a dated copy. Openness made this discrepancy visible.

The archive is also more careful than a single headline number. It separates ASNs, IPv4 prefixes and IPv6 prefixes. It groups by year and, for address resources, by prefix length. Other literal statuses include RECLAIMED, ADMIN and, in some files, REVOKED. A separate series counts organisations by year. These distinctions matter because a voluntary return, a membership closure and a revocation are not interchangeable descriptions of institutional action.

AFRINIC's own resource-management guidance gives reclamation an operational meaning. It describes de-registration from the AFRINIC Whois database, including related child records such as reverse-DNS, routing, RPKI and DS material. It also says reclaimed resources pass through quarantine and sanitisation before possible reissue. A chart label is therefore several layers away from the act it purports to summarise.

The right response to a visible label defect is not to discredit the whole registry record. It is to keep those layers separate and repair the public one with evidence.

What the 2026 rows actually say

The current files report this narrow picture:

Resource class VOLUNTARY in 2026 VOLUNTART in 2026 Other 2026 observations
ASN 3 1 1 RECLAIMED
IPv4 4 1 2 RECLAIMED
IPv6 0 1 1 RECLAIMED

The IPv4 VOLUNTARY count is distributed across /22, /18 and /23 groups. Its VOLUNTART row is a /22. The IPv6 row with the second spelling is a /32; the other 2026 IPv6 observation is a reclaimed /48. The ASN file has no prefix-length field for its corresponding row, as one would expect.

These are aggregate observations, not address quantities. Adding them across resource classes would create a number with no coherent unit: an ASN is not an IPv4 prefix, and neither is an IPv6 prefix. Nor may the three suspect rows be treated as one transaction. The separate organisation file says nine organisations were closed in 2026, but it exposes no event key connecting an organisation to any of the resource-class rows.

The frozen CSV and JSON representations match as complete tuples of resource type, status, prefix length, year and total. That removes one tempting explanation. The second spelling is not confined to a stale download while the web chart uses corrected data, or vice versa. Both public representations carry it.

Exact strings become institutional categories

The dashboard code makes the consequence concrete. For ASN, IPv4 and IPv6, it creates a separate dimension from the t_status field and groups the exact values before summing each total. No checked client-side rule maps VOLUNTART to VOLUNTARY.

That design is ordinary and often desirable. Exact values preserve differences instead of silently blending them. Here, however, exactness turns one character into statistical identity. A chart or query that selects only VOLUNTARY will produce three ASN observations, four IPv4 observations and no IPv6 observation for 2026. A reader who decides that the other token is a misspelling will produce four, five and one.

The latter interpretation is linguistically persuasive. It is not yet authorised by the source. The public material reviewed for this article contains no status dictionary saying what VOLUNTART means and no correction notice mapping it to another value.

This is why “just fix the typo” is an incomplete governance answer. If the file is overwritten, future readers may obtain a cleaner total without knowing that an earlier public version differed. If every analyst normalises the word independently, the record remains reproducible only inside each analyst's private assumption. If nobody normalises it, a likely accidental category acquires a life of its own.

The choice is not between untidy truth and tidy censorship. A registry can preserve the original observation and publish an accountable correction.

What the evidence cannot carry

The public aggregate does not name a member, holder, number resource or closure case. It does not state who entered the status, when during 2026 the event occurred, which notice or agreement applied, or whether the same underlying case produced all three resource-class rows. It cannot establish that any holder voluntarily surrendered resources.

It also cannot show that a registry action was unlawful, inaccurate or technically incomplete. There is no evidence here of a routing change, reverse-DNS failure, RPKI error, service interruption or loss of rights. An underlying event could be fully authorised and correctly executed while its statistical label is misspelled. The inverse is also possible in principle: perfect spelling cannot prove a sound decision.

The public data layer should therefore be judged for what it controls. AFRINIC controls the generation and publication of the files and the dashboard's interpretation of them. The applicable registry and corporate processes control a closure, return or reclamation. Resource holders control voluntary requests within those processes. Network operators control their running networks. A status label reports an asserted disposition; it does not create one.

Correct the category without rewriting the past

A durable repair needs two things that a replaced file alone cannot provide: a stable vocabulary and a correction lineage.

The vocabulary should give each disposition a machine-stable code independent of its display spelling. It should define voluntary return, member closure or reclamation, administrative adjustment and revocation, including the aggregation unit used for each resource class. Display labels can then be corrected without inventing a new status.

The lineage should preserve the source snapshot, generator version and dataset hash; the raw token; the authorised mapping, if any; the affected years, prefix groups and counts; the correction time and responsible function; the old and new hashes; and a clear statement about whether the change touched only the public projection or any underlying registry record.

None of that requires publishing a member's confidential closure file. Aggregate correction can remain aggregate. The purpose is not to expose private evidence but to make the public assertion traceable.

If AFRINIC confirms that VOLUNTART was meant to be VOLUNTARY, the result is a narrow projection repair. If it defines the token as a genuinely different state, the dictionary should explain the difference. If the token simply disappears from a later snapshot, the current chart will become easier to read but the historical question will remain unless the versions are linked.

Three rows and one character do not amount to an institutional scandal. They do reveal whether a public ledger treats correction as part of truth or as housekeeping performed out of sight. AFRINIC has already supplied the more important half of the answer by publishing the data. It should complete the record by showing how the category changes.

Sources