Summary

  • APNIC’s 2025 Activity Plan set an outcome of 50 new RIPE Atlas probes deployed in the region.
  • The 2025 Annual Report records 60 probes distributed and 23 deployed in the corresponding result cell, plus two Atlas anchors deployed as a separate outcome.
  • Twenty-three is 46% of the 50-probe target, but 23 divided by 60 is not an observed conversion rate: the public documents do not establish that both counts describe the same cohort.
  • A privacy-safe lifecycle receipt should reconcile distribution, registration, first connection, 30- and 90-day availability, disconnection and retirement. Without it, a logistics count cannot be traced into durable measurement capacity.

There is a large operational distance between putting a probe in someone’s hand and gaining a vantage point on the Internet. APNIC’s annual reporting gives both ends of that journey a number. It does not give the journey an identity.

The 2025 Activity Plan described the work plainly: deploy RIPE Atlas infrastructure to improve Internet measurement. The target outcome was 50 new probes deployed in the region. The Annual Report returned to the same row and retained that target, then reported 60 probes distributed and 23 deployed. Beneath that line it recorded two RIPE Atlas anchors deployed.

These are useful disclosures. They show that distribution exceeded the target count as a physical handoff measure and that APNIC distinguished some devices as deployed. They also expose a measurement problem. “Distributed” and “deployed” are different units of work. A table that places them together without a shared cohort leaves the reader unable to reconcile activity with outcome.

The clean comparison is 23 against 50 because both figures are labelled deployed. On that basis, the reported result is 46% of the target, with 27 deployments not reconciled in the row. That does not mean 27 devices were lost. It does not mean APNIC failed to support their hosts. It means the public outcome does not say whether those deployments moved into another period, remained pending, came from an earlier distribution batch or were counted by a rule not stated in the report.

The tempting comparison is 23 against 60. It yields 38.3%, but it is not a measured conversion rate unless the 23 deployed probes are known to be a subset of the 60 distributed probes. The reviewed documents do not make that join. Dividing the numbers anyway would turn a missing data relationship into a false precision.

The annual report changes the unit mid-row

Targets work only when their units survive contact with the result. A target of 50 deployed probes can be met, missed or deferred. A result of 60 distributed probes answers a different question: how many devices left the programme’s custody or entered a handoff process. The second measure can explain effort, reach or future capacity, but it cannot replace the first.

Distribution is a flow at a logistics boundary. Deployment is a state, although even that word needs a definition. Does a device count as deployed when a host accepts it, when it receives power and Ethernet, when it is registered, when it connects for the first time, or after it stays usable long enough to contribute measurements? Each rule produces a different total at the same reporting date.

The Annual Report does not publish the rule in that row. It also does not say whether “60 distributed” is a 2025 flow, a cumulative inventory or a mixture that includes replacements. The 23 deployed figure may refer to the same devices, to a subset, to devices distributed earlier, or to all probes reaching a programme milestone during the year. Those possibilities are not accusations. They are incompatible accounting models.

The two anchors make the need for stable units clearer. RIPE Atlas anchors are more capable measurement systems and targets; the report lists two separately. Adding them to the 23 probes would produce 25, but that would mix device classes and still not satisfy a 50-probe target. The report is right to keep the classes apart. It should apply the same discipline to lifecycle stages.

RIPE Atlas already has the states needed for reconciliation

The missing evidence does not require a new theory of deployment. RIPE Atlas’s public documentation already distinguishes the transitions.

Its probe API records when a probe first connected and when it last connected. It carries a current status, the time that status changed and accumulated uptime. The documented lifecycle includes Never Connected, Connected, Disconnected, Abandoned and Written Off. Where disclosure is permitted, the record can also describe country, ASN, prefixes, tags and firmware version.

Those fields matter because a powered device is not automatically a connected probe. RIPE Atlas defines connected as having a working Internet connection and maintaining a connection to the Atlas infrastructure. A local network can be functioning while a firewall or path problem leaves the probe classified as disconnected. Installation and measurement availability are therefore related, but not identical.

The troubleshooting guide draws an even earlier boundary. It has specific advice for probes that ambassadors have given out but hosts have not yet plugged in. It separately addresses probes that have never connected. In other words, the operating system itself recognizes a chain: handoff, installation, registration, first connection and continuing service.

APNIC need not reproduce private RIPE NCC records or identify volunteers. It can use an opaque cohort key and aggregate the transitions. The public question is not where a particular probe sits. It is how many opportunities handed to hosts became stable, diverse measurement points under a declared cut-off rule.

A cohort is the denominator the programme is missing

“Cohort” can sound like statistical decoration. Here it is simply the identity of a batch. If 60 devices were distributed during 2025, give that set a public, non-personal label. Then report how many were registered, first connected, remained connected for 30 days, remained available for 90 days, disconnected, were replaced or had not yet activated by year end.

The same ledger should show carry-over. A probe distributed in December may connect in January without representing failure. A probe distributed in 2024 may first connect in 2025 and legitimately contribute to the year’s deployed total. Without a cohort and transition dates, both events vanish into the same pair of annual numbers.

Cut-off rules are equally important. A “deployed” count taken on 31 December is a stock. A count of probes that first connected at any point during the year is a flow. A count of devices connected on at least 90% of days is an availability measure. All three can be useful, but they cannot share one label without changing the story.

The receipt should also preserve losses without assigning blame. A host may change providers, lose power, replace a router, encounter a firewall, move premises or simply decide it cannot continue. RIPE Atlas warns that individual disconnections do not by themselves measure an ISP’s quality. Aggregated reason codes and state changes can improve programme operations without turning volunteers into performance targets.

Measurement capacity is not a hardware total

APNIC’s own 2026 plan raises the stakes. It promises to report Internet-performance changes in newly established IXP economies using RIPE Atlas data. That is a sensible measurement method. It also means that the quality of the result depends on the vantage points available before and after an intervention.

A global APNIC Blog article reported 13,421 connected probes as of February 2026 and described a daily connected population ranging from roughly 13,300 to 14,400 in the second half of 2025. It also found 34 economies with only one probe or anchor on the selected day. These are global observations, not results for APNIC’s 2025 distribution programme. They nevertheless show why one extra stable point in a sparse economy can matter more than several boxes sent into an already dense network.

Placement is one dimension. Persistence is another. A first connection proves that installation crossed the controller boundary. It does not prove a year of usable observations. A probe connected for 90 days contributes a different evidential asset from one that appeared briefly and disappeared, even though both can be called deployed under a first-connection rule.

Connectivity class matters too. RIPE Atlas reporting distinguishes public and non-public probes and records IPv4 and IPv6 connectivity where available. A programme seeking better regional measurement may care whether a new vantage point adds an economy, an ASN, an access-network type or an address family. A raw count does not tell members which blind spot changed.

None of this invalidates measurements already made. A latency result can be technically sound without a public programme ledger. The governance issue is traceability: when an annual plan funds distribution and a later plan relies on Atlas evidence, members should be able to follow the conversion from effort to durable capacity without guessing which count is the denominator.

A compact receipt would close the chain

The minimum public record is small. It needs an opaque cohort ID, reporting period and cut-off date. For each cohort it should publish aggregate counts for devices assigned, distributed, registered, first connected, connected for at least 30 and 90 days, disconnected, abandoned, written off and replaced. Probes and anchors should remain separate.

It should state whether late activations roll into the next year and whether replacements appear in both distribution and deployment counts. Broad economy and ASN-class totals can be published only where privacy and re-identification risks permit. Exact addresses, host names and precise locations are unnecessary.

The receipt should reconcile to the target. If the 23 deployed devices came from the 60 distributed in 2025, say so. If some came from an earlier cohort, show that transfer. If the other devices were pending at cut-off, give the aggregate pending state. If the word deployed means first connected rather than currently connected, define it once and retain the definition across years.

This would turn the annual table from a pair of achievements into a controlled conversion record. It would let APNIC distinguish a procurement or distribution success from an activation constraint, focus support where hosts actually get stuck and preserve credit for devices that come online after the reporting boundary.

What the evidence permits

The sources support a narrow conclusion. APNIC targeted 50 new deployed probes for 2025. It reported 60 distributed and 23 deployed, with two anchors separately deployed. RIPE Atlas provides distinct lifecycle and connection fields that could support an aggregate reconciliation. The reviewed APNIC documents do not publish that reconciliation.

They do not show that 37 of the 60 devices failed. They do not identify a missing device, an inactive host, a particular economy or a cause. They do not establish that 23 belongs inside 60, so no actual distribution-to-deployment conversion percentage can be claimed. The global Atlas counts do not fill the gap.

The finding is about units. APNIC counted motion at two points in a lifecycle but did not give the points a common cohort. Until it does, 60 describes reach, 23 describes a state, and neither number explains how much durable measurement capacity the programme created.

The box is not the vantage point. The receipt is what proves the transition.

Sources