UK Public Sector GitHub Observatory
Who actually runs UK government's websites. Every fact on this page comes from a DNS record the operator published — the CNAME naming a supplier, the addresses a domain resolves to, and the network announcing them. Nothing here connected to a server, requested a path or tested a configuration.
The estate
Before any of the numbers
6,054 of 7,913 domains. They are the reason every naive count on this page would be wrong: there are thousands of them, they are the smallest tier of local government, most have no IT department and several have no staff at all. What they buy is shared hosting, and left in the same pot they outvote every department in the country. Every chart below is split accordingly.
Suppliers: departments, agencies and principal councils
1,859 domains — everything that is not a parish, town or community council.
Domains by identifiable supplier
From the CNAME where there is one - a department's own DNS naming its supplier - and from the announcing network otherwise. Orange bars are CDNs and proxies: they answer for the domain without revealing what is behind it, so those counts are a floor.
Suppliers: parish, town and community councils
6,054 domains, and a different market entirely.
Domains by identifiable supplier, smallest tier only
Commodity shared hosting, website builders and one very large CDN. This is what the .gov.uk estate looks like by volume, and almost none of it is what people mean when they say 'government hosting'.
Networks
Domains by announcing network, parish tier excluded
Autonomous system numbers, mapped through Team Cymru's public DNS interface. Colour is where the AS is registered, which is not where the servers are.
- UK-registered network
- registered elsewhere
The top five account for 4,645 of 7,199, and the top twenty for 6,197. Concentration is not automatically a problem — a small number of large, well-run platforms is usually better than three hundred small ones — but it is a fact worth knowing before an outage makes it obvious. Nothing here says whether any two of these networks share a physical dependency.
Documentation sites, design systems and prototypes, published on GitHub Pages under 166 distinct subdomains rather than on a service.gov.uk address. They are real government-facing sites, they are outside the .gov.uk estate this page probes, and they are therefore outside whatever monitoring covers it.
The sites government publishes about itself
Separate measurement, separate source: these are the homepages set by hand on 2,572 of 25,626 harvested repositories. Nobody sets that field by accident, so every entry is a site somebody wanted people to find.
Most-referenced websites, by repositories pointing at them
Purple bars are GitHub Pages subdomains.
GitHub Pages subdomains
Documentation sites, design systems and prototypes, published on GitHub Pages under 166 distinct subdomains rather than on a service.gov.uk address. They are real government-facing sites, they are outside the .gov.uk estate this page probes, and they are therefore outside whatever monitoring covers it.
Method: CNAME, A and AS lookups over the same 7,913 .gov.uk domains
as the domain security page, run
2026-08-28 in 306s.
6 of those records were carried forward from an earlier sweep, because this run learned nothing about them: two full sweeps close together get throttled, and a throttled lookup is indistinguishable from a domain that does not resolve. A run that loses domains and gains none is being refused, not observing a change, so it is never allowed to overwrite a better answer. Names
beginning with an underscore are excluded — _dmarc,
_acme-challenge and the hashed names certificate authorities ask
for are DNS plumbing rather than websites. AS mapping by
Team Cymru, used through its public DNS interface and
gratefully. Supplier classification is a rule list over CNAME suffixes and AS
names: it identifies what it recognises and says nothing about the rest, which
is why "90% on an identifiable supplier" is a
statement about the rule list as much as about the estate.
See Method.