Tukuk-OSFree Is Not the Same as Open
A PDF of a national census buried on a ministry website is free — you can look at it. It is not open: you cannot copy it, machine-read it, or build on it. The definition of open data crystallised in the 2000s–2010s — from the Berlin Declaration on Open Access to Knowledge to the Open Knowledge Foundation's tidy "Open Definition" — into two pillars: legally open — an explicit license allowing reuse, ideally CC0/CC-BY, with no registration — and technically open — bulk access, stable URLs, structured formats (CSV, JSON, GeoJSON), documented columns. Miss either half and you have a publicity photo of data, not data.
The second pillar is the one that decides whether anything gets built. NASA's API returns JSON in a documented shape; that is why hundreds of sky apps exist. A ministry that "publishes" statistics as interactive dashboard widgets technically shows numbers and legally forbids the download — the same numbers, zero ecosystem.
Who Actually Keeps It Alive
The romantic answer — governments — is half right. National statistics offices are open data's anchor tenants: census bureaus, meteorological services (NOAA's feeds are the reason every weather app works), geological surveys (USGS earthquake feeds, no key, ever), EU's Eurostat. Then a second tier does the work governments structurally cannot: community mappers (OpenStreetMap, 10 million contributors, the base layer of half the "commercial" maps you use), citizen scientists (GBIF's biodiversity records, EONET's near-Earth events), library-and-museum digitisers (Met Open Access, Project Gutenberg, Europeana), and NGOs that scrape, tidy and republish what was technically free but practically unusable.
Open data persists through institutions with mandates, not enthusiasm: USGS has a legal duty to publish; a hobby mirror of USGS dies when the hobbyist moves. When evaluating a source, the useful question is boring: who is contractually or legally obliged to keep this URL alive for the next decade?
The Graveyard Problem — and How to Spot a Living Source
Most dead open data is not deleted; it just stops updating. Three tells: (1) the dataset's "last updated" field lags its own stated cadence; (2) the bulk endpoint works while the pretty website is broken (or the reverse — a sign nobody is watching); (3) the issue tracker or mailing list has unanswered reports older than a year. Favourite trap: an API key you must request — a free key is still a gate, and gated sources rot quietly because the maintainers can count registered users and feel successful while actual use collapses. The honest maintenance metric of open data is reuse: who shipped what, this quarter, on top of it.
Your fallback should be archival thinking: the Wayback Machine, official mirrors, and a personal copy of any table your project depends on. Data licensed for reuse can be cached; that right is the entire point of the license.
Try It Yourself
We run the whole world section of this site on keyless public data — the free-data API roundup lists the workhorses (REST Countries, World Bank, USGS, Wikipedia) with their honest fine print. Pick one number you care about — your country's population, the strongest quake this month — and trace it to its primary open source rather than an aggregator; then check that source's update stamp. You will have performed the exact audit that decides whether a dataset is infrastructure or a museum exhibit — and you will understand why our pages cache aggressively, label archives honestly, and cite the maintainer by name.