It started out as a crowdsourced spreadsheet:

That spreadsheet held about 25 cities. That was a great start, but crowdsourced data has by nature challenges with consistency. For example, some people have a more expensive taste than others and will tell you the rent in a city is higher than the actual average.

To mitigate this, we've since switched to public data sets by the UN, WHO, IMF, World Bank for things like demographic, crime, safety and healthcare information, as well as public APIs for things like weather, air quality and traffic density. All data is processed and normalized constantly. Practically that means there might be 42 different samples for air quality in Amsterdam, and 9 different internet speed measurements for Tokyo. Our robots will remove outliers, discount older values and calculate median values that have the highest probability of being accurate when you arrive in a place. The more data (and thus samples) is inserted into the site, the more accurate the overall data becomes.