Summary

  • A study on the RIPE 93 agenda found 731 probes, 2.4% of those examined, with at least one physically implausible reported location between January 2019 and April 2026.
  • Researchers referred 258 currently flagged probes to RIPE Atlas. By April 2026, 103 hosts had answered its notices: 93 changed the location and ten contested the flag.

The RIPE 93 listing schedules Katherine Izhikevich to explain the work on Tuesday evening. The more revealing part of the original paper is a handoff. Researchers find a map pin that cannot be reconciled with network timing. RIPE Atlas sends a notice. The person who controls the probe decides whether to move the pin.

The test uses measurements Atlas already collects to DNS root servers. A probe reports where its host placed it on the map. The researchers compare its fastest observed round trip with the minimum time light in fibre would need to reach a root-server instance from that claimed location. They require the contradiction to appear against two independently operated roots and allow a 100-kilometre buffer. That produces a conservative flag for the host to check and correct.

Across the study period, 731 probes were flagged at least once. In the April 2026 snapshot, 155 were flagged. Those numbers describe different windows. The correction campaign sent 258 live cases to RIPE Atlas between July 2025 and April 2026. Of 103 replying hosts, 93 accepted and updated their location; ten said the pin was right. Four of the ten dissenting hosts said the probe ran on rented infrastructure whose physical location they did not control directly. A server lease is a surprisingly poor substitute for a tape measure.

RIPE Atlas's own guide gives the host or manager the edit button. Other users can dispute a badly placed probe; the host can move the pin or confirm it. The researchers publish a changing list of likely bad probe IDs for measurement users to check. Their historical analysis also found that half of eventual host updates came after at least thirteen weeks of implausible measurements.

That delay travels downstream. The paper's re-analysis of one RIPE IPmap method found 28 likely mislocated probes among roughly 2,800 used in a year of measurements, touching 1,125 target IPs that may have been assigned the wrong city. A small wrong dot can cast a long shadow when a measurement treats it as ground truth.

The practical question is wonderfully ordinary: when a probe's position is disputed, can the next person using its measurement see the date of the flag and the host's answer? The map pin has an owner. The evidence should have a receipt.