Data cleanup: our lovely routine

When I started Geographic History, I had no ambition to make its data the best. Now I simply have to - and I love it :)

As a matter of fact, a history app has one job before any other: be right. If the map says a city belonged to someone it never belonged to, that might feel like a small bug - but to you it reads as a fact.

Believe it or not, every source online has errors - even the best ones, like Oxford's curated databases, and even the ones built by large communities over many years. In 2026 there's AI slop, but human errors came long before that.

So day by day, we keep cleaning it up. We have a daily fact-checking routine, and for many cases it works like this:

1. We pick a period. Not something obscure like 400-600 (that's another story), but a more or less well-documented one, like the 19th century.
2. We run several automated scans using scripts and a couple of mighty LLMs, Claude Opus-level ones.
3. They report the most likely errors and pass them on to fact-checking. That part can be done by smaller LLMs, and it takes most of the time.
4. After searching online and bringing back sources, the fact-check produces a list of map patches.
5. Then comes the fully human part: rechecking those patches, reviewing possible mistakes in them (even though there are very few), and applying them (or sending them back for corrections).

Some numbers behind it:

1. We have more than 25,000 region-years. That is, more than 25,000 facts to check.
2. In September and October alone, more than 600 data errors were found and corrected.
3. More than 1,000 suspicious spots are still being researched, so expect more corrections soon.

Sounds overwhelming and boring, right? Not for us :) The result justifies the work. After applying hundreds of these patches, we're quite sure we now have the best source of geohistorical data in the world. Still, this job isn't done - and it hardly ever will be. New archaeological finds, Wikipedia edits, historical texts newly translated and uncovered - all of it gives us a reason to keep doing this daily. And, as I said, I really love it :)

P.S. We've also added a way to report mistakes: there's now a "Spot a mistake?" button right in the app. Every report gets checked the same way.