Summary

  • A LACNIC Blog contributor says the University of Los Andes is conducting a RIPE Atlas study of how the 24 June submarine-cable cut affected national latency and response times. The public account does not identify a measurement ID, probe cohort, target, observation window, raw result or analysis method.
  • Cirion’s public sequence separates work begun on 19 July, fibre fusion and end-to-end tests on 22 July, and a complete-reconnection claim on 23 July. The later contributor account uses 24 July for confirmation of full repair and says local fibre, radio-site and power problems continued.
  • RIPE Atlas currently lists 109 Venezuelan probe records, 44 connected at the time checked, across 26 recorded IPv4 ASNs. That September inventory is not the June event cohort and cannot be used as an affected-user denominator.
  • A useful study record must join a retrievable measurement object to source-labelled event states. Temporal association can select a phase for investigation; it cannot, by itself, name the cable, route, power or access transition that caused a latency change.

Three dates can describe one repair without describing the same state

The Cirion update of 22 July is unusually useful because it stops before the easy word “recovered”. Specialists aboard the Wave Sentinel, it says, had located and prepared both ends of the cable and completed the individual fusion of the optical fibres. End-to-end tests showed connectivity from Venezuela to international connection points in Panama and St. Croix within the original design specifications.

The same update also says what remained. Encapsulation would be followed by pre-immersion tests that night, before international optical verification. A joined fibre and a satisfactory test are therefore milestones in a sequence. Neither sentence establishes that the cable had been returned to ordinary service, that traffic had shifted back from alternatives, or that users beyond the landing system had recovered their earlier experience.

On 23 July, Cirion described a larger state. The operator said hospitals, businesses and families were connected again and called the result a complete reconnection 29 days after the earthquake. It also said alternative international routes had kept more than 60% of its customers operative from the first days, that the Caracas ring had been restored in under 48 hours, and that 15 government authorisations had been obtained in 14 days.

Those are operator statements, not independent measurements of every Venezuelan connection. The customer denominator is not defined on the page. “Operative” is not resolved into reachability, capacity, latency or service class. Yet the update is still a valid event source: it records when the cable operator publicly assigned a complete-reconnection state to its work.

The LACNIC Blog contribution published on 19 August uses 24 July for confirmation that the South American Crossing cable was fully repaired. It then says that local fibre and radio base stations still needed repair and that some areas lacked electricity or reliable power. The article expressly notes that its author’s views do not necessarily reflect LACNIC’s views.

One date can mark a passed instrument test. Another can mark an operator’s service decision. A third can mark a later author’s confirmation. None automatically stands for the restoration of every access network. The chronology becomes misleading only when the state labels disappear.

The event began with two clocks 32 seconds apart

The public USGS event query gives the 24 June shock a more precise starting boundary than a calendar date. Event us6000t7zc, magnitude 7.2 in the captured record, has an origin time of 22:04:31.806 UTC. Event us6000t7zp, magnitude 7.5, follows at 22:05:04.053 UTC. The interval is 32 seconds.

That record does not establish which shock damaged the cable, or even establish the physical failure mechanism. The contributor account places the cut near the coast and near the second epicentre; that remains the author’s account. The USGS identifiers are useful for a narrower reason: “after the earthquake” is not an adequate start time for a study that may compare changes measured in seconds or minutes.

Other transitions are less precise in the public material. The contributor says terrestrial routes to Colombia were activated over the following weeks. Cirion’s 19 July update says the repair vessel had entered Venezuelan waters, the two broken ends had been found, and integrity and connectivity tests were in progress. It does not turn every intermediate routing change into a timestamped record.

A measurement window must therefore carry the quality of its event boundary. A USGS origin time, an operator publication time, an asserted effective time and a later retrospective date are different fields. Converting them into one neat vertical line would create precision rather than preserve it.

The announced study has a subject but not yet a public measurement object

The LACNIC-hosted contribution says: at the University of Los Andes, “we are conducting a study using RIPE Atlas probes” to measure how the cable cut affected national latency and response times. That sentence supplies a research subject and an intended method family. It does not supply the object a reader would need to inspect.

The public account does not name a RIPE Atlas measurement ID. It does not say whether the measurement is ping, traceroute, DNS or another type; which target represents the relevant traffic path; whether IPv4 and IPv6 are separated; which probes were selected; when the baseline, impaired, rerouted and repaired windows begin and end; or how disconnected probes and missing results are treated. No raw-result URL, aggregation rule, route-path comparison, code version or uncertainty estimate is linked.

This is not evidence that the study lacks those things. It is not evidence that a fuller method or result exists nowhere else. It means only that the public announcement cannot yet be followed from claim to measurement.

That distinction is the article’s boundary. A research team may be doing careful work before publication. A public reader may still be unable to reproduce a public description. Reporting the second fact does not justify inventing the first.

Forty-four connected probes are a present state, not a June cohort

The RIPE Atlas probe API offered a tempting number when checked: 109 records associated with Venezuela, 44 then marked Connected, carrying 26 distinct integer IPv4 ASN values. It would be easy to put 44 beneath a chart and call it the country sample.

That would be wrong.

A probe’s current status says nothing by itself about whether it was connected on 24 June, whether it participated in the announced measurement, whether it produced complete results through a cable impairment, or whether its path crossed SAC. A probe is not a household, a user, a mobile site or a unit of traffic. Several probes can sit in one network; one probe can change status during the event; a disconnected probe can be part of the very missingness the study must explain.

The cohort has to be reconstructed at event time. Its record should include the exact probe IDs, selection rule, first and last connection state relevant to the window, ASN and address family, useful geographic resolution and every exclusion. If a probe disappears during the impaired phase, the analysis must decide whether that absence is a network observation, a probe failure or an unknown—not silently remove it from the denominator.

Current inventory remains useful. It shows that public probe metadata exists and that present visibility spans more than one IPv4 network. It cannot travel backward in time and become the experiment.

RIPE Atlas already has an address for inspectable results

The RIPE Atlas results documentation makes the missing identity concrete. Full result history is retrieved below a measurement ID. The endpoint accepts lower and upper UNIX time limits and a comma-separated filter of probe IDs. Results can be large, so the service streams them.

A compact publication need not reproduce every packet. It can name the measurement, preserve the exact query used for each phase, record the retrieval time and hash an export. A reader can then check which probes returned data and whether a later correction changed the dataset.

There is a local precedent for that discipline. The 2024 ACM SIGCOMM paper Ten years of the Venezuelan crisis — An Internet perspective names Atlas measurement 1591146 for platform-wide traceroutes to Google Public DNS. It states a 30-minute cadence, analyses the first five days of each month, computes the minimum RTT per probe in a monthly snapshot and discusses restricted platform coverage and partial visibility.

That paper is not evidence about the 2026 earthquakes or the SAC repair. Its old measurement cannot be borrowed as the new study’s dataset. It demonstrates something more modest: a Venezuela latency analysis can publish the identity, cadence, selection and limitations that make its result inspectable.

Put the event ledger beside the latency series

The minimum record contains two objects.

The measurement object names the ID and visibility state; owner; creation time; type; target; interval; address family; start and stop; selected probes and event-time status; baseline and comparison windows; raw-result export; missingness classes; aggregation; code version; and correction history. If the interpretation depends on rerouting, route-path observations belong there too.

The event record names a source, asset, geography, effective time and state. The state should be specific: earthquake observed, cable impairment reported, alternative route in use, repair work started, fibre fused, end-to-end optical test passed, pre-immersion work completed, cable returned to service, local access restored, or power restored. It should say whether the evidence is an operator assertion, instrument result, regulator record, independent observation or later account.

The join is deliberately thin. It says that these result rows fall within this documented event state. It does not promote proximity in time into causation. If latency falls after an operator’s reconnection statement, the record may identify a recovery-phase association. Naming the cable as the cause requires evidence that the measured paths used it, that the route changed as asserted, and that plausible local access, power and target changes were considered.

This leaves the decision where the evidence lives. A network operator can run a local test, move bounded traffic, inspect a path or restore a preferred route without waiting for a central verdict. The change should have an observation window and a rollback condition. A public study supplies common evidence; it does not become the operator of the network it observes.

Sources