UK Public Sector GitHub Observatory
Every public repository in github.com/MHRA was mirror-cloned in full and
mined from its git history rather than the GitHub API alone. This site reports what 8 repositories, 9k commits and 34 people
actually did.
The estate at a glance
What stands out
3 of 8 repositories are read-only. That is not neglect so much as a decade and a half of deliberate turnover: GDS has consistently retired work rather than leaving it to rot in place. The live estate is the 5 repositories that remain, and only those are a fair basis for judging current practice.
They are not archived, so they still read as supported, but 40% of the active estate has had no push in over two years. This is the gap between 'archived' and 'actually maintained' - the population where dependency drift accumulates unobserved.
Shape of the estate
Repository status
Archived repositories are read-only. GDS archives aggressively, which is healthier than the alternative but makes the raw repository count a poor measure of the live system.
- Active, own code
- Active fork of a third party
- Archived (read-only)
Repositories created each year
The 2017 peak coincides with the GaaP era - Notify, Pay, Registers - and a wave of departmental prototyping.
Year of last push
A decay curve. The 2026 spike is the live estate; everything left of it has stopped moving.
Volume of work
Commits per year, from git history
Measured from the commits themselves across every branch and tag, not from GitHub's activity graph.
The same history, month by month
Monthly resolution shows the texture the yearly view flattens: the long climb to the mid-2010s peak, the plateau, and the bot-driven noise of the Dependabot era. The busiest single month was 2020-04 with 786 commits.
Languages
Primary language, by repository count
GitHub's headline language per repository. Ruby is the house language of GOV.UK publishing; Python dominates the data and alerting estate.
What GDS built
GDS has never been one product. Repository names carry the programme they were built for, and those prefixes outlive the programmes themselves – which makes them the clearest surviving record of what the organisation worked on, and when it stopped.
Repositories per programme
Classified by naming convention. 'Other / experimental' is the long tail of prototypes, spikes and one-off tools.
GitHub topics, as set by GDS itself
An independent check on the classification above: topics are applied by hand, and 0 of 8 repositories (0%) carry at least one. The ranking closely tracks the name-derived programmes, which is reassuring for both.
Commits per year, by programme
The rise and fall of each programme, measured in actual commits rather than announcements.
- Forked third-party
- Other / experimental
- Infrastructure & platform
Programme life-spans
The period covering the middle 98% of each programme's commits. Deliberately not first-to-last commit: Emergency Alerts was built from the Notify codebase and inherited a stray 2011 commit with it, which made both programmes appear to start a decade before they existed. Trimming the tails ignores inherited and vendored history and answers the question actually being asked – when was this being worked on.