Light in fiber travels about 200 km per millisecond. That sets a hard floor on how fast this browser can reach any given city — and no VPN, proxy, or edited location setting can move that floor. Run the probe and watch the physics triangulate you.
How it reads. Each probe opens a connection to a server in a known city and times the round trip. Not all of that time is flight: about 15 ms goes to connection setup and the server answering, and the fiber between two cities runs roughly 1.35× longer than the straight line between them. Take those out, halve what's left, and multiply by 200 km/ms — that is one constraint: you are inside that circle of that city. The shaded lens on the globe is the intersection of every constraint so far — it starts as the whole planet and shrinks with each reply until it locks onto the one region all eight round trips agree on. Congestion and detours only ever make a link look longer, so a short round trip is proof of nearness that cannot be faked in software.
How precise it gets. The solver itself is exact — given clean timings it lands on the true point. What limits it is jitter: 1 ms of timing noise is 100 km of range. In practice a stable connection resolves to a few tens of kilometres, a noisy one to a few hundred. The figure quoted after the coordinates is this run's own uncertainty, measured by how far the answer can move before it stops explaining the readings — not a fixed guess.
When it flags. A mismatch is only declared when a single round trip excludes the location on file — when that city is farther from a server than the reply could possibly reach even granting zero connection overhead, the most generous reading physics allows. That test is one-sided and cannot be argued with, because congestion can only add time. Everything short of it is reported as a question: the best fit landing somewhere else is weak evidence, since a detoured route drags the fit without disproving anything, and a run whose timings allow half the planet is reported as inconclusive rather than as a match.
What it can't do. Estimates land within a few tens of kilometres at best, and routing that detours (satellite links, corporate tunnels, congested peering) inflates every reading and pushes the guess outward. The 15 ms of overhead and the 1.35× detour factor are typical values, not measurements of your particular path — a slow server or an unusually straight route moves the rings either way. Treat a mismatch as a question, never an answer.