Paste the same address into three different lookup tools and you can get three different cities. People usually read that as two of them being broken. It is not. It is what happens when several parties independently guess at something nobody actually measures, and understanding why is the difference between distrusting all IP geolocation and knowing when to trust it.
Nothing measures where an address is
This is the part that surprises people, so it is worth being blunt about. An IP address carries no location. It is a routing number. There is no GPS in it, no field that says Warsaw, and no authority that stamps one on. Every city you have ever seen beside an address is an inference, and it comes from a small number of places:
- Registry records. Regional registries record which organisation a block of addresses was allocated to, and where that organisation is. That is a corporate headquarters, not a customer. A national provider registered in the capital will drag its entire country toward that city.
- What the provider publishes. Providers can publish a file mapping their own blocks to places, which is the only source in this list that comes from someone who genuinely knows. Many publish nothing. Many publish something and never update it.
- Network timing. Measuring how long an address takes to answer from several known points puts bounds on where it can physically be. Good for continents, weak for cities.
- People telling them. Submitted corrections, and location data collected from devices. This is where most street level precision actually comes from, and it is the least visible part.
Why two locators show different cities
Once you see the list above, the disagreement stops being mysterious. Two tools showing different cities are usually not contradicting each other about a fact. They are weighting those sources differently, or they bought their data from vendors who did.
A few specific things make it worse in ways that matter on a video chat platform. Providers reassign blocks between regions and the old assertion lingers. Mobile networks route through regional gateways, so a phone can be genuinely in one city and appear in another hundreds of kilometres away. And an address behind a VPN resolves to the server, correctly, which looks like an error unless you know to expect it. We went through those failure modes in more detail in why the location is sometimes wrong.
What none of this gives you is a way to fix a bad reading. You can notice it is wrong. Under the model above, the only people who can correct it are the vendors, and they will not hear from you.
Corrections belong to the block, not the address
The fix has to work at the right level of detail, and the right level is not a single address. Home addresses are reassigned constantly, so a correction filed against one is about a customer who may be gone by next week.
What is stable is the block: the range of neighbouring addresses a provider hands out together. A provider assigns a block to a region and the block stays there while individual addresses inside it move between customers. So a correction filed against the block outlives the connection that prompted it, and it applies to every future connection from the same range rather than to one address once.
That is also what makes correcting worth anybody’s time. Fixing one address helps nobody. Fixing a block quietly fixes every reading from it afterwards, including for people who never noticed it was wrong.
Why one person saying so is not enough
Here is the obvious problem with letting people correct a map: people are wrong, people guess, and a small number of people will deliberately poison it. A system that serves whatever the last person typed is worse than the vendor data it replaced.
So a correction does not become an answer on its own. It becomes a vote. A place is only served once several members have independently filed the same one and that place holds a clear majority of the votes on that block. Until both are true, nothing is published for it.
The interesting state is the third one. When members disagree and no place clears the bar, the block is marked contested, and the honest thing to do with a contested block is not to pick a winner. So the panel shows the competing places with their shares instead, something like Warsaw 48%, Krakow 31%, Katowice 21%, and lets you read the uncertainty rather than hiding it behind a confident single answer.
None of this is a claim you have to take our word for. The counts are published on the platform report: how many blocks have been corrected, how many cleared the gate and are actually being served, how many are still contested, and the breakdown by country. The gate itself is printed next to the numbers so the figures can be read rather than trusted.
What this still cannot fix
A community layer improves the sources underneath it. It does not change what the whole method is capable of, and the ceiling is worth stating rather than stepping around. Reliable at country level, good but not perfect at city level. It resolves the network a connection is registered to, never a street address, a house number or a device.
- A VPN is not an error to be corrected. The reading is right, it is just about a server. Correcting it would make the data worse, not better.
- Mobile blocks resist it. A carrier gateway serving half a country has no single correct city, so there is no right answer for members to converge on.
- Coverage is uneven and always will be. Blocks get corrected where people actually connect. A country our members rarely meet has few corrections, which is exactly why the report publishes the per country breakdown instead of one accuracy percentage that would hide it.
- It never gets more precise than a neighbourhood. No amount of agreement turns a network block into a street address, and anything claiming otherwise is selling something.
The short version
- Nothing measures where an address is. Every IP location is an inherited claim from a registry record, a provider file, network timing, or people submitting data.
- Locators disagree because they weight those sources differently, not because some of them are broken.
- Corrections have to be filed against the block rather than the address, because addresses move between customers and blocks do not.
- One person is not enough. A corrected place is served only when several members independently agree and one place holds a clear majority, and blocks where they do not agree are shown as contested with the competing shares.
- The counts, including per country, are published on the platform report so the claim can be checked instead of believed.