Server-location counts are the one headline figure a provider can raise without touching its network, you edit a list. So we treat the count as a vendor claim and check it against a thing no provider controls: how fast light moves through glass.
Two dimensions sit on opposite sides of this. Server locations is vendor-claimed: it is the provider’s own list, and a flag next to a hostname costs nothing to print. Server authenticity is measured, it is what checks the list. The check is not clever, and that is its strength. It is arithmetic on a constant.
The constant
Light in optical fibre travels at roughly 200 km per millisecond, about two-thirds of its vacuum speed, slowed by the refractive index of the glass. Nothing carrying your packet beats that. So a measured round-trip time (RTT) to a server puts a hard ceiling on its distance from us:
200 km/ms. Queuing, routing, and processing only ever add time, so this is a floor on nothing and a ceiling on distance: a small RTT cannot hide a large distance.A worked example
Put a probe in Singapore and ping a server the list labels “Tokyo.” Singapore to Tokyo is about 5, 300 km in a straight line. Light in fibre needs roughly 26.5 ms to cross that one way, so an honest round trip cannot come back in much under 53 ms, and real submarine cable routes are longer than the straight line, so the true figure is higher still.
Now suppose the probe measures an RTT of 8 ms. Halve it for one way, multiply by the constant: 4 ms × 200 km/ms = 800 km. Whatever answered is at most 800 km from Singapore. It cannot be in Tokyo. It is somewhere in or near Singapore, wearing Tokyo’s label, a virtual location, counted in a number sold as global reach.
The reasoning only runs one way, which is what makes it safe. A large RTT is ambiguous, a server really could be far away, or the route could simply be congested. But a too-small RTT is dispositive: no amount of good engineering lets a packet arrive before light could. To place a location rather than merely refute one, we triangulate from several probe vantage points and intersect the distance ceilings each one implies. The result is a bound the provider cannot argue with, because arguing with it means arguing with the speed of light.
Where this stands today
This is the method, stated ahead of the data. The current corpus holds no server-authenticity measurements yet, so the dimension carries no score for anyone, it is excluded from the weighted mean rather than guessed at. When the probe network is live it will be the one dimension we produce entirely ourselves, which is why its ceiling is a full 10: unlike a speed test we cite from others or a streaming claim we read off a page, a distance bound is a number we can reproduce on demand. It is the dataset this site intends to own.