How we score cities

CrossBorders rates 283 cities across 14 categories built from 109 individual metrics — 30,000+datapoints in total. Every datapoint carries a source, a confidence level, and the date it was last checked. Nothing on a city page is generated on the fly: the same dataset produces the same scores and the same prose for everyone, and city prose is written from that data, then human-verified — we never invent claims a number can't back.

Three kinds of metrics

Objective metrics are measured facts: rent for a two-bedroom flat, homicide rate, summer heat, internet speed. Subjective metricsare traits no agency publishes — walkability of the central neighbourhoods, how secular public life feels, café culture. We score these on written rubrics with fixed anchor cities, so “7/10 walkability” means the same thing in Osaka as in Porto. Derived metricsare computed from other data plus your own answers, such such as an open-market one-bedroom as a share of one typical employee's net pay, or distance from your family.

Where the data comes from

Objective values primarily come from public and global datasets, each applied the same way to every city so the numbers stay comparable. Country-level indicators — governance and political stability, physicians per capita, unemployment — come from the World Bank; homicide statistics from UNODC; tax burden from OECD Taxing Wages and published country tax summaries; weather from long-run climate normals and reanalysis-derived series. City population, density and local salary levels come from national statistical offices (the US Census Bureau, INSEE, Destatis, ABS, StatCan, IBGE and their counterparts). Comparable urban-centre population comes from the European Commission Joint Research Centre's GHS Urban Centre Database 2025 under CC BY 4.0 and the release's peer-reviewed methodology. We extract the matching centre and round its population to whole people. For six entries without a coherent centre, we identify and label an explicit alternative geography, such as a locality, functional area, combined statistical area, agglomeration or city-state catchment. These CrossBorders changes are not endorsed by the Commission. Rents come from official sources that record actual rental transactions — lease and deposit registers, compulsory contract registrations, and national housing statistics — and from CrossBorders market observations where those sources do not reach (see the next section). Day-to-day costs, internet speed and air quality come from global datasets that publish one consistent methodology across hundreds of cities — we take their definition rather than stitching local ones together, because a number measured the same way everywhere is what makes a comparison mean anything. These examples are not an exhaustive source list.

What “rent” means here

Our rent figures are open-market rents: what you would typically pay signing a new lease on an ordinary flat, measured across the whole city — not just prime neighbourhoods, and deliberately not the average of what current residents pay. Those are different numbers. Long-standing tenancies, regulated, subsidised and informal housing mean the average resident often pays less — and in strongly segmented markets, far less. In Cape Town, for example, the median renting household pays under R2,000 a month by official survey, while an advertised one-bedroom runs several times that; both numbers are true, and only the second one describes what a newcomer will actually face. Since that is the question our users are asking, that is the number we show.

The separate rent-to-income ratiodeliberately combines that newcomer rent with one local employee's mean net pay. It asks whether a typical employee could rent a whole new-lease one-bedroom alone; it is not the share that current households actually spend. A high value often means people share, live with family, use social or regulated housing, or earn above the mean. Where the calculated share runs beyond the stored ceiling, we say it is beyond typical local pay rather than printing a precise-looking capped percentage.

Wherever possible the figures come from sources that observe real market transactions: statutory deposit and bond registries, compulsory lease registrations, and national statistics built on new contracts. Where no such source reaches the market, we may retain a dated CrossBorders observation of advertised rents; its exact acquisition evidence stays in our internal audit trail while the public source label remains generic. Where a direct observation is not available, values are CrossBorders estimates based on economic data and market observations and carry a lower confidence label — estimated, not measured.

Rubric-scored traits are estimated with AI assistance against those written rubrics, then calibrated: every metric has founder-reviewed anchor cities, and each value stores a comparison to its anchors (“≈ Berlin, above Cape Town”) so estimates stay consistent across the whole dataset rather than drifting city by city. We label every datapoint high, medium, or lowconfidence and keep that label honest — roughly one in nine datapoints is marked low, and we'd rather show you that than pretend to precision we don't have.

How scores are normalised

Every metric maps to a 0–100 score in the direction that helps a resident: cheaper rent scores higher, more crime scores lower. Some metrics aren't “more is better” at all — January temperature and city size follow utility curves with a sweet spot — and some, like traditionalism or career intensity, are neutral traits whose value depends entirely on what youwant; the engine decides their direction from your answers, not from our taste. Category scores on a city page are the plain average of that category's measurable metrics and are identical for every visitor.

Personalised fit

Your ranking from the free quiz weights the same objective scores by what matters to you. Two rules keep it honest. First, dealbreakers are genuine hard floors only— things like high crime, extreme heat, weak healthcare, or needing to live car-free. A dealbreaker penalises a city only if you selected it, the penalty grows smoothly with how badly the city fails (no cliff where 60 km flips a verdict), and soft preferences like walkability or rent are never double-counted as both a weight and a penalty. Second, trade-offs are surfaced, not hidden: if your #1 city is expensive or far from family, the results say so.

What we don't do

No city pays to rank higher. No affiliate deal reorders a list. No score is tweaked for a headline. And when the data is thin, the confidence label says so instead of the page pretending otherwise. Where we show accommodation offers, they are chosen by the city you are already looking at — never the other way round, and no provider can buy a position in any ranking.

Freshness and corrections

Each datapoint stores when it was last checked; most of the dataset was reviewed in 2026. Cities change, and a 30,000+-point dataset will contain mistakes — if you live in one of these cities and a number looks wrong, tell us via the support pageand we'll check it against sources and fix it.