Summary

  • APNIC’s December 2024 announcement bounded its preliminary IDNIC work: all 2,974 allocations made from 1 January 2019 through 19 April 2024 were reviewed, and 1,216 IPv4 delegations shared data points that required more investigation. The 1,216 were questions, not findings.
  • APNIC separately reported a 51-record IRINN question set, with 44 delegations later found not to meet policy guidelines and recovered. Its completed JPNIC, TWNIC and KRNIC reviews, however, have no public reviewed-record denominator in the frozen program record.
  • Public Executive Council minutes describe a potential Second Opinion Review gate using the most recent 20 delegations after a systemic-non-compliance threshold. The threshold is not public, and 20 must not be called random, representative or equivalent to a ten-year review.
  • A versioned, privacy-safe completion receipt could bind population, policy and procedure versions, method, sampling limits, aggregate findings and dispositions, sign-off, uncertainty and corrections without exposing Member applications.

A status is not yet evidence. It is an index entry pointing toward evidence.

This distinction matters because APNIC’s Resource Delegation Review Program now has three NIRs carrying the strongest available status word: completed. The 6 July 2026 update says the TWNIC and KRNIC analyses were fully completed, their minor queries and clarifications resolved, and they joined JPNIC. Read in isolation, that sounds decisive. Read against APNIC’s own earlier disclosures, it is strangely unmeasured.

We know exactly how many records one preliminary IDNIC exercise examined. We know the dates defining that population. We know the size of the preliminary question set. We know the opening and closing numbers of a separate IRINN investigation. We do not know the published population behind any of the three completed-review labels.

That gap is not evidence of a hidden problem. It is evidence that the word “completed” is doing more work than the public artifact beneath it.

The strongest defence is operational, not rhetorical

The review program answers a real obligation. A registry cannot maintain an accurate ledger merely by accepting that every historical delegation was correctly assessed and will remain correctly recorded forever. APNIC says the program’s primary activity is to review all IPv4 delegations and transfers made by APNIC and each National Internet Registry over roughly ten years. False, stale or policy-inconsistent entries can weaken trust in the registry long before they create a visible routing incident.

The institutional design also makes the work difficult. The APNIC region has seven existing NIRs: APJII/IDNIC, CNNIC, IRINN, JPNIC, KISA/KRNIC, TWNIC and VNNIC. They are not APNIC branch offices. They are separate organizations operating under local law, language, evidence and procedure while applying regional and global obligations. A field that looks anomalous to an external analyst may be explained by an authorized local workflow. A neat cross-registry test can therefore produce a false positive when it ignores the context in which an allocation was made.

APNIC’s public language recognizes this limit. When its December 2024 review found common data points across 1,216 IDNIC delegations, it did not call them improper. It called them preliminary questions and referred the work to APJII, which held information APNIC could not access. That was the correct boundary. A signal is not a finding, and the institution nearest the underlying evidence must have a chance to explain it.

Confidentiality is part of the same defence. Member applications can contain business plans, contracts, customer information, identity records and security-sensitive material. Publishing those files would not make the program more accountable; it would make participation more hazardous. A public completion standard must therefore work with aggregate evidence, controlled suppression and explicit limits.

Clean outcomes should be defended too. A ten-year review can finish with no adverse findings. Minor queries can be resolved without becoming breaches. Local interpretation can remove false positives. Accountability fails when every clean conclusion is treated as suspicious, just as it fails when every status word is accepted without a boundary.

The case for a completion receipt begins by accepting all of these points. It asks for a better public record, not a more accusatory review.

Where APNIC already publishes a denominator

The 17 December 2024 announcement is a useful model precisely because it states what its numbers mean.

APNIC reviewed all 2,974 IDNIC allocations made from 1 January 2019 through 19 April 2024. Within that bounded population, 1,216 IPv4 delegations shared data points that raised preliminary questions requiring more investigation. The date range matters. It prevents the 2,974 from being misdescribed as IDNIC’s entire history or as the later full ten-year review population. The word preliminary matters. It prevents the 1,216 from being turned into 1,216 allegations.

The announcement also describes a different funnel at IRINN. An investigation started with 51 questionable delegations made in 2022. IRINN later confirmed that 44 did not meet policy guidelines and recovered them. Here the public can see a progression from a question set to an aggregate disposition. It cannot inspect the Member files, nor should it. It can nevertheless distinguish the population under question from the outcome.

That is the basic virtue of a denominator. It lets the reader separate three statements that otherwise collapse into one another:

Public statement What it establishes What it does not establish
2,974 allocations reviewed in a fixed window The bounded preliminary IDNIC population The later ten-year IDNIC review population
1,216 shared data points Records needing further investigation Fraud, breach or invalidity
44 of 51 recovered by IRINN An aggregate investigated disposition The identity or private evidence of any Member

The subsequent IDNIC figures reinforce the need for dated states. The June 2025 update reported 104 delegations identified for reclamation. The September update reported 102 confirmed for reclamation, 257 designated as meeting policy and 318 Second Opinion Review requests processed since July 2024.

The difference between 104 and 102 should not be repaired by imagination. The sources describe two dated stages using different verbs. Records may have been reclassified, corrected or otherwise processed, but the public texts checked here do not supply a reconciliation. The honest account preserves both numbers, their dates and their labels. A correction history in a completion receipt would make the movement intelligible without requiring an outsider to invent the bridge.

Three completed states without three completion tables

JPNIC offers the cleanest contrast. APNIC’s February 2025 update says its ten-year analysis of JPNIC registry data began in January, completed in February and had no adverse findings. This is a legitimate conclusion. The missing fact is elementary: how many delegations and transfers were in the reviewed population?

The public update does not say. Nor does it attach a table identifying included record classes, exclusions, controlling policy versions, method changes or aggregate queries. The result may be sound. It is simply not reproducible as a bounded public claim.

TWNIC and KRNIC arrived at the same status by a different route. In December 2025, APNIC said the TWNIC data analysis was complete while minor queries and clarifications were still being resolved. The same update said randomized APNIC spot checks had produced no adverse findings, but it did not publish the accumulated sample size. By July 2026, APNIC said both TWNIC and KRNIC were fully completed and that the minor queries and clarifications had been resolved.

Again, none of this implies an undisclosed adverse case. The point is comparative. APNIC’s program page provides a useful quarterly index and links to posts, presentations and recordings. Across the frozen index, updates, conference material, Annual Report and Activity Plan, there is no separate JPNIC, TWNIC or KRNIC result report with a reviewed population or a per-review completion table.

A program can therefore answer “where are we?” but not “what public evidence does this completed state summarize?” Those are different questions.

The July 2026 map makes the distinction more important as the queue grows. VNNIC and IDNIC were in result evaluation. IRINN and CNNIC were in initial review. APNIC’s own registry review was planned to begin in the third quarter of 2026. Without a stable receipt format, each quarterly status adds another noun to the program history but no common unit for comparing closures.

Local context belongs inside the method

The VNNIC update of October 2025 supplies affirmative evidence for caution. APNIC and VNNIC jointly interpreted records against local regulation and operational practice. That work identified false positives that were being corrected in the review sample.

This is not an embarrassment to be hidden. It is what a careful review should do. A method that cannot change when local evidence contradicts an initial inference is not rigorous; it is merely consistent. The relevant accountability question is whether the correction is recorded.

A completion receipt should therefore identify both the regional policy version and the authorized local-procedure version applied to each period. It should state whether the analysis method changed, which data-source classes were used and how earlier flags were corrected. The aim is not to make local procedure subordinate to a regional template. It is to show how the two were joined.

This also protects NIRs. Without a method record, a public count can become a reputational label detached from the reasons why it changed. With a versioned receipt, a false positive can be recognized as a corrected analytical state rather than quietly disappearing or remaining forever as an insinuation.

The most recent 20 are a gate, not a denominator

The most specific public description of a Second Opinion Review gate appears in the APNIC Executive Council’s September 2025 public minutes.

The sequence is conditional. Potential breaches must reach a level indicating potentially systemic non-compliance. If that level is reached, APNIC checks the full application data for the NIR’s most recent 20 delegations. A Second Opinion Review is applied only if that sample shows inconsistent policy application.

Three limits must stay attached to this disclosure.

First, the numerical systemic trigger is not public in the checked record. It cannot be reconstructed from the number 20. Second, “most recent” is a selection rule, not a synonym for random or statistically representative. Recent records may be especially relevant to current procedure, but they do not automatically describe the preceding decade. Third, the minutes do not establish that this gate was used for JPNIC, TWNIC or KRNIC. Joining the gate to those three reviews would be invention.

The gate may still be sensible. APNIC said it could not handle more than two remediation processes, including two SORs, at the same time alongside its own resource-delegation work. Scarce review capacity requires sequencing. A bounded gate can keep the most intensive process proportionate.

But proportionality is only publicly testable when the gate has semantics. A receipt could describe the threshold in exact terms or, if disclosure would enable gaming or expose confidential enforcement logic, give a bounded explanation of why the value is withheld. It could state that the frame was the latest 20, why that frame was chosen, what it can reveal and what it cannot.

The number 20 should not become a magic seal. It is a decision instrument whose limits belong in the record.

What a completion receipt would contain

The missing artifact is smaller than a report and more useful than another status post. It is a versioned, per-NIR completion receipt.

It should contain twelve fields:

  1. the review target, the institutional roles and the authorities signing the closure;
  2. the data cutoff, review window and record classes;
  3. the total population, included and excluded counts, and reasons for exclusions;
  4. the controlling regional-policy versions and authorized local-procedure versions by period;
  5. the analysis method and version, data-source classes and later corrections to the method;
  6. the systemic-trigger definition, or a bounded explanation of why its exact value cannot safely be public;
  7. every sample frame, selection rule, size, purpose and stated limit;
  8. aggregate flags, queries, clarifications, false positives and unresolved cases;
  9. aggregate dispositions such as compliant, corrected, reclaimed, pending, escalated or otherwise closed;
  10. the remediation and SOR state, including the evidence used to exit either process;
  11. residual uncertainty, the confidential-evidence boundary and any suppressed small cells;
  12. the completion date, review owner, NIR acknowledgement, APNIC sign-off and correction history.

Small groups need not be published as exact cells. Counts can be banded or suppressed where they could identify a Member. Sensitive trigger values can be described by function rather than disclosed numerically when APNIC records the reason. The receipt should never include application files, customer names, business plans, personal data or security-sensitive evidence.

This is not radical transparency. It is disciplined minimalism. The public learns the shape of the review and the meaning of its outcome. The institution retains the confidential evidence needed to reach it.

Completion should be portable

APNIC’s program already contains the components of a stronger public record. It has bounded populations in the IDNIC announcement. It has an investigated funnel at IRINN. It has dated disposition changes. It has evidence that local-context review corrects false positives. It has a conditional SOR gate and an explicit capacity constraint. It has quarterly status reporting.

What it lacks is a common closing object that joins those components for each NIR.

The purpose of that object is portability. A reader encountering “JPNIC completed” in a quarterly post should be able to follow a durable reference and learn the population, method, limits and closure. A later correction should update the history without rewriting the earlier record. A new APNIC staff member should not have to reconstruct the conclusion from slides, minutes and blog posts. An NIR should not have to defend a clean result with private material merely because the public status is too thin.

There is an old institutional temptation to treat a clean review as needing less documentation than an adverse one. The opposite is often true. An adverse action creates its own paper trail through remediation, appeal and recovery. A clean closure can disappear into a word. That makes positive evidence the least portable part of the system.

The right standard is symmetrical. Preliminary suspicion should be bounded so it does not become accusation. Adverse disposition should be aggregated so it can be understood without exposing a Member. Clean completion should be measured so it does not become faith.

APNIC does not need to prove that every allocation was wise. It needs to say what population it reviewed, under what rules, by what method, with what corrections, and what “completed” excludes. That is enough.

A registry is a bookkeeper. The highest form of bookkeeping is not the confident stamp. It is the receipt that lets another reader understand exactly what the stamp covers.