Summary

  • RIPE NCC’s CAPIF 5 report says RIPE Atlas measurements from 2022, 2024 and 2026 suggest more direct and intra-regional paths in Central Asia, fewer long external detours and less reliance on Russia.
  • A 2024 RIPE NCC presentation documents a largely comparable method, but it also shows a changing probe population: 90 probes in four countries versus 91 in 2022, with substantial changes inside the country totals.
  • The same presentation calls the result qualitative rather than quantitative and lists important limits, including unknown traffic weight, incomplete probe coverage, geolocation uncertainty, tunnels, ECMP and path instability.
  • A versioned sample ledger should link each country-pair finding to measurement IDs, time windows, probe cohorts, successful and failed traces, unknown-hop handling, protocol and address-family counts, classification rules and result denominators.

The strongest line in RIPE NCC’s new Central Asia report is not the conclusion that paths have improved. It is the smaller warning placed beside that conclusion: more diversity does not automatically mean more resilience.

That distinction keeps an observed route from becoming a claim about the whole network. A traceroute can show a path seen from one vantage point to one destination at one time. It does not weigh that path by traffic volume, contracted capacity, commercial importance or the number of users behind it. It does not reveal every layer-1 or layer-2 segment. It does not prove why the path changed.

The CAPIF 5 report, published on 17 September, is generally careful about those boundaries. It says RIPE Atlas measurements carried out in 2022, 2024 and 2026 suggest more direct routes within Central Asia, fewer long detours outside the region and a reduced—though still significant—reliance on Russia. It then gives three country-pair examples.

Between Kazakhstan and Uzbekistan, a significant proportion of observed 2026 paths is described as direct. Between Kyrgyzstan and Tajikistan, the report contrasts a 2022 set in which most measured traceroutes did not reach the destination with a 2026 picture containing a significant number of direct and shorter intra-regional paths. Between Kyrgyzstan and Uzbekistan, it reports more direct routes and significantly lower delays. In the diagrams, ?? marks hops that could not be geolocated.

Those are useful observations. They are also exactly the kind of observations whose future value depends on keeping the denominator attached.

The method has a public history

The comparison did not appear without methodological context. In a September 2024 CAPIF 3 presentation, Alex Semenyaka said that the 2024 research repeated the 2022 work using the same methodology, with RIPE RIS correlation added.

The slides describe the construction. Sources were all RIPE Atlas probes in one country. Destinations were Atlas probes in other Central Asian countries, plus additional hosts for Turkmenistan. For every source and destination, the method used UDP, TCP and ICMP over IPv4 and IPv6. Hops were associated with countries, converted into country chains, aggregated by source and destination countries, and compared with control-plane evidence from RIS.

The presentation also supplies a fact that should travel with every later comparison: the cohort moved. It lists 90 probes in four countries in 2024, against 91 in 2022. Kazakhstan fell from 63 to 53 and Tajikistan from nine to eight; Kyrgyzstan rose from six to eight and Uzbekistan from 13 to 21. The headline totals are almost identical, but the distribution of vantage points is not.

That does not invalidate the work. A live measurement platform cannot freeze the Internet or its volunteers. It does mean that “same methodology” and “same sample” are different propositions.

The same deck is unusually direct about bias. Not every network prefix has an Atlas probe. The real traffic weight of an observed route is unknown. Layer-1 and layer-2 geography can remain hidden, especially through multinational operators and tunnels. Router geolocation may be absent or wrong. ECMP may still appear imperfectly even with Paris traceroute. Routes can change during one measurement and over time. The stated conclusion is therefore a qualitative picture, not a quantitative one.

That disclosure is a strength. A sample ledger would carry it forward rather than replace it.

“Significant” needs a count, not an accusation

The CAPIF 5 article uses “significant” in ordinary editorial language. It does not label a statistical test, confidence interval or significance threshold. Nothing in the reviewed material justifies reading the word as a formal statistical claim.

The problem is more practical. A reader cannot tell from the article how many successful traces sit behind “a significant proportion”, how many failed to arrive, how many contained unlocated hops, or whether one path class represents a large set of reporting probes or a handful of combinations. Nor can the reader see whether the 2026 classifier treated loops, repeated countries, private addresses and unknown hops exactly as the 2022 and 2024 implementations did.

The answer is not to remove the adjective and pretend uncertainty has vanished. It is to publish the counts that let the adjective remain modest.

For each year and ordered country pair, the receipt could name the public RIPE Atlas measurement IDs and observation window. It could list requested, scheduled and reporting probes; source and destination cohorts; results by UDP, TCP and ICMP and by IPv4 and IPv6; successful, failed and incomplete traces; and the count of hops that were private, ambiguous or not geolocatable.

The next block would describe the transformation: country-geolocation source and snapshot, rules for repeated countries and loops, definitions of direct, intra-regional, Russia, other out-of-region and unknown paths, latency filter and statistic, plus a notebook or classifier version. A small table would then show the numerator and denominator for every category displayed in the diagram.

RIPE Atlas already gives measurements durable identifiers. Its API documentation distinguishes a measurement object from a result returned by one probe, exposes measurement settings and timing, and allows results to be requested by measurement ID, probe ID and time window. The proposed receipt is therefore not a new measurement platform. It is a join between public identifiers and the published conclusion.

What the receipt must not become

A provenance record can be useful without becoming a permission system. RIPE NCC should remain free to choose research questions, instruments and safe disclosure. Operators remain free to choose peers, transit providers and route policy. Atlas probe hosts contribute vantage points; they do not vote on what every network should do.

The record also need not expose private target addresses, unpublished probe locations, credentials, commercial agreements or security-sensitive topology. Aggregate counts, public measurement IDs, versioned rules and result hashes are enough to establish continuity.

Nor would a receipt prove cause. The CAPIF 5 article discusses IXPs, new physical links, regional operators and policy developments. Those may be relevant context. A before-and-after path diagram does not, by itself, allocate causal credit among them. The right public statement remains: these were the observed paths under this cohort and method.

That sentence is less dramatic than “the region changed because of X”. It is also more reusable.

A comparison that can survive its own success

Central Asia’s measurement surface should improve. More Atlas probes, more direct peering and more successful traces would all be welcome. But improvement creates a historical problem: the better the future sample becomes, the harder it is to remember what the earlier diagrams could actually see.

A versioned ledger preserves both states. If a geolocation error is corrected, a classifier changes or a late result is admitted, the new receipt should supersede the old one without erasing it. If 2028 has twice as many usable source-destination combinations as 2026, the comparison can state that directly instead of presenting a smoother line than the evidence permits.

RIPE NCC’s published conclusion may well remain the same after that work: more direct and intra-regional paths, with important dependencies still present. The value of the receipt is not that it must reverse the conclusion. It is that an operator, researcher or policymaker can see what kind of conclusion it is.

A path diagram is a view. A sample ledger is the memory that lets views become a series.

Sources

  1. RIPE Labs — CAPIF 5: Regional Interconnectivity in Central Asia
  2. CAPIF 5 — Interconnectivity in Central Asia: 2022, 2024, 2026
  3. RIPE NCC — CAPIF 3 presentation: Interconnectivity in Central Asia
  4. RIPE Atlas documentation — Measurements
  5. RIPE Atlas documentation — Measurement results endpoint
  6. RIPE NCC — CAPIF 5 agenda