Every dataset on this site, what produced it, when we fetched it, how much of the site it actually covers, and where it falls short. The last of those is the reason this page exists: a number you cannot check should say so, and 6 of our 14 datasets are our own judgement rather than a measurement.
Taken straight from whoever produces or governs the data. If you follow the link you land on the same number.
SourceOpen-Meteo Historical Weather API (ERA5 reanalysis)
Retrieved6 August 2026
HowDaily 2019-2023 normals per city coordinate, averaged to monthly highs, lows and precipitation by scripts/build_city_climate.cjs.
710 of 710 cities
SourceIANA Time Zone Database
Retrieved6 August 2026
HowZone resolved from each city's coordinates with tz-lookup, then evaluated at runtime by the Intl API so DST is always current.
710 of 710 cities
SourceWikimedia Commons file pages
Retrieved6 August 2026
HowPhotographer, licence and file page recorded per image in images/cities/attribution.json and shown as a visible credit on the page. Only PD, CC0, CC BY and CC BY-SA accepted.
713 images
SourceNumbeo cost-of-living index
Retrieved5 August 2026
HowComponent prices scraped per city in local currency exactly as Numbeo reports them, converted to USD at apply time. Numbeo is itself crowd-sourced, so it is primary only for the Numbeo index, not for ground truth.
331 of 710 cities
Numbeo sample sizes are small in less-visited cities and its 'last updated' date varies widely per city.
SourceOpen-Meteo Elevation API (Copernicus DEM GLO-90)
Retrieved9 August 2026
HowElevation sampled at each city's stored coordinate. Not published; used by scripts/check_data_sanity.cjs to catch island and region centroids that land on high ground.
710 of 710 cities
SourceThe provider's own website, or an embassy list of physicians published by language
Corroborated againstbangkok.diplo.de/th-de/service/aerzteliste-1506722, Each provider's own site, recorded per row as sourceUrl in data/service-languages.json
Retrieved12 August 2026
How scripts/build_services.cjs refuses to build if any row lacks a sourceUrl, a known evidence tier, a known language or a real city slug. The tiers are printed on every card. scripts/verify_service_links.cjs re-fetches every url and sourceUrl and reports status, redirect target and page title, so a link that rots or starts pointing somewhere else is caught rather than assumed; the 2026-08-12 runs removed one row whose cited article did not contain the claim, dropped two invented domains, corrected a link that redirected to the wrong branch, and later dropped five more rows whose sources turned out to be a dead domain, a 404, an unreachable directory, or a site that never actually states a spoken language. The checker retries three times before calling a link dead, because a single transient failure once reported six healthy URLs as dead. One live site (vetinlisbon.pt) refuses Node's fetch entirely and was confirmed by hand.Every row records the exact URL the language claim was read on and an evidence tier: official (a government or embassy list defined by working language), self-declared (the provider's own site), directory (a third-party aggregator only), or visited (we confirmed in person; unused so far). Read the rest
253 providers across 21 of 710 cities, 12 service types and 32 languages. 6 official, 131 self-declared, 116 directory. Chiang Mai 34, Tbilisi 30, Lisbon 19, Bangkok 18, Hanoi 16, Da Nang 16, Medellin 14, Ho Chi Minh City 14, Barcelona 11, Mexico City 11, Bali 10, Budapest 10, Belgrade 8, Dubai 8, Buenos Aires 7, Prague 6, Berlin 5, Warsaw 5, Cape Town 4, Kuala Lumpur 4, Tallinn 4. Coverage is 3 percent of the site's cities.
An owner decision to grow volume via aggregator directories took the directory tier from 39 rows to 105, roughly half the set, so the average row is now weaker than when the dataset was 62 rows. Many directory rows rest on a ROSTER-LEVEL claim ("the clinics we list speak English") rather than a statement about the named business; each such row says so in its note. PRIMARY FOR THE CLAIM, NOT FOR FLUENCY. We source that a provider is stated to work in a language; we have not verified that it does, and no row carries the 'visited' tier. A hospital offering interpretation is counted under the languages it interprets, which is not the same as a doctor speaking them: Chiang Mai Ram and Thonglor Pet are interpreter desks, not multilingual clinicians. The 17 directory-tier rows rest on guides and aggregators that may sell placement, making them the weakest entries. Only Bangkok has official-tier rows. The US Consulate Chiang Mai health-facilities and legal-provider PDFs and the US Embassy Georgia mental-health resource PDF are all image-only or protected and could not be read; retrieving those would upgrade several rows and is the obvious next improvement. A Russian-language directory of Russian-speaking doctors in Georgia (needdoc.ru) is an unexploited language-defined source of the same kind. Coverage is a pilot, not a map of any city's real provision, so absence from this list means nothing.
No usable primary source exists, so the figure is agreed across at least two independent sources, both named here.
SourceWikipedia 'Visa policy of X' cross-checked against official government portals
Corroborated againsten.wikipedia.org/wiki/Category:Visa_policy_by_country, Official e-visa and immigration portals of the destination countries (evisacuba.cu, elsalvadorpei.migracion.gob.sv, snedai.com, and equivalents)
Retrieved8 August 2026
HowDefault rule established from the article lead first, exceptions layered on top, then every country sanity-checked against five known passports (DE/US/UK/IN/NG) before acceptance.
15 of 117 destinations verified this way; the original 102 destinations predate this rule and their origin was never recorded
UNVERIFIED BACKLOG: the 102 older destinations are not sourced. Mongolia's visa-free programme expires 2027-01-01 and Myanmar's visa-on-arrival entries are dated trials, so both need re-checking.
SourceGoogle Maps ratings, mostly via aggregators that republish them, plus OpenStreetMap for location
Corroborated againstwww.google.com/maps, nominatim.openstreetmap.org
Retrieved6 August 2026
HowVenues gated at a 4.0 published rating, then geo-checked against OpenStreetMap within 25km of the city centre. 65% resolve to an OSM record, 15% carry a verified official website, 20% rest on the research pass alone. Coworking spaces may instead be admitted on confirmed current operation without a rating.
713 venue cards across 710 cities
Most ratings came from aggregators mirroring Google rather than Google directly, because the search and scraper budgets ran out. A handful in the Sendai and Tashkent sets were conservative estimates rather than read values.
Our judgement, not a measurement. Listed here in full because a number you cannot check should say so.
SourceOur own assessment
How13 category scores assigned per city by editorial judgement. No measurement, no external index, no sample.
710 of 710 cities
It is currently presented as if it were measured. Replacing it with sourced inputs (Numbeo crime index for safety, Ookla for internet, IQAir for air quality, Numbeo for cost) is the single highest-value provenance fix outstanding.
SourceNumbeo for 331 cities, our own estimate for the rest
HowWhere Numbeo data exists the headline figure is reconciled against it; otherwise it is an editorial estimate.
331 of 710 sourced, 379 estimated
The 379 estimated figures are indistinguishable from the sourced ones on the page.
SourceWritten from model knowledge, venue and neighbourhood names verified by search
HowSeven sections per city. Venue names, districts and visa mechanics were checked during writing; the price ranges and internet speeds quoted in the prose were not individually sourced.
61 cities have researched long-form sections; the remainder use the older generated text
Contains specific figures (rent bands, coworking day passes, Mbps) that carry no citation.
SourceWritten from model knowledge, district names verified by search
How295 entries across 59 cities researched in 2026-08. Coordinates hand-placed from known landmarks and validated against the city centre, not geocoded.
710 of 710 cities have a section; 295 entries are from the 2026-08 pass
Coordinates are accurate to roughly a few hundred metres, fine for a map pin but not authoritative. Price levels are relative judgements.
SourceHand-authored
HowCurrency, language, mains voltage, tap water guidance, tipping norm, ride-hailing apps and emergency numbers written per country without a cited source.
All countries in the dataset
Emergency numbers and tap-water guidance are safety-relevant and should be moved to a primary source (ITU and each country's civil protection agency).
SourceHand-authored
HowPlug type letters per country, written without a cited source.
All countries in the dataset
IEC World Plugs is the obvious primary source and is not yet used.
Tiers are defined in our provenance manifest, which this page is generated from, so the two cannot drift apart. Primary means taken directly from the body that produces or governs the data (a met agency, IANA, a government visa portal, the photographer's licence). Triangulated means no usable primary source, so agreed across >=2 independent secondary sources, each listed in corroboration. Editorial means our own judgement. Not a fact. Must be labelled as opinion wherever it is shown.
Found a figure that looks wrong, or a source we should be using instead? Tell us on the contact page. Corrections are the cheapest way this page gets better.