Geofeeds and geolocation

Working out where an IP address physically is has no single authoritative answer, so Atlas collects several and compares them. When the sources disagree — which is often — the disagreement itself is the useful signal.

The sources, strongest first

Geofeeds (RFC 8805)

A geofeed is a CSV an operator publishes at a well-known URL, mapping their own netblocks to country, region and city:

192.0.2.0/24,US,US-CA,Mountain View,94043
2001:db8::/32,DE,DE-BE,Berlin,

This is the strongest IP-geolocation source Atlas has, for the obvious reason: it is the operator stating where their own addresses are, in a machine-readable form they maintain. Lookups do a longest-prefix match across the collected feeds.

Atlas maintains a curated list of feed URLs, refreshed daily. A weekly discovery scan finds more from geofeed: attributes in whois and queues them disabled for review, so an unvetted feed never silently starts influencing results.

Feeds are re-fetched with ETag and Last-Modified, so a stable feed costs nothing to check. Prefixes an operator removes disappear on the next refresh rather than lingering.

Commercial geolocation databases

Broad coverage, no publisher accountability, and quality that varies a lot by region. Useful where nothing better exists; not something to override a geofeed with.

RIPE's probe coordinates

What the probe host declared when registering. Unverified and often approximate — sometimes deliberately, since hosts may not want their home pinpointed.

Measurement-derived constraints

Latency puts an upper bound on distance: light in fibre only goes so fast, so a 5 ms RTT rules out a location on another continent. This can contradict a claim but never confirm one — many places are within 5 ms of you.

Atlas uses this as one link in the chain when judging whether an anchor really is in the country its record says.

Cross-checking

The IP lookup page carries a geo cross-check card that puts the sources side by side. Read agreement as reasonable confidence and disagreement as a question, not as one source being wrong — a data-centre range legitimately geolocates differently depending on whether you mean where the hardware is or where the service is presented.

Geofeed accuracy (/network/geofeed-accuracy, staff) scores published feeds against the other evidence — the view for "is this operator's feed any good?"

Place names

Separately from addresses, Atlas needs stable names for places. It uses three vocabularies:

• Country codes — ISO two-letter, used everywhere as the coarse key
• GeoNames — the city and place-name backbone
• UN/LOCODE — stable codes for locations, which disambiguate the many cities that share a name

Which one applies depends on the view; see the Locations guide.

Reporting a wrong location

Geolocation is where Atlas is most often visibly wrong, and it is exactly what Issues are for. A flag on a specific record with the evidence attached is far more actionable than a general complaint.