UK Public Sector GitHub Observatory
13 August 2026 · deletions method transparency
There is a hole in the middle of this project, and it is worth naming. Every score, every percentage and every finding on this site is computed from repositories that are public today. An organisation can improve its numbers by deleting the code that drags them down, and a weekly harvest of live repositories would never know.
A prototype with no licence, no README and a key in its history is a finding right up until somebody removes it, at which point it becomes nothing at all. That is not a hypothetical failure mode. It is the ordinary consequence of measuring only what is still there.
Somebody is already keeping the receipts
github.com/uk-gov-mirror is an unofficial mirror of every UK government repository on GitHub, maintained by Jonty Wareing, updated at least weekly, and now standing at over 25,000 repositories. It exists precisely so that removals are recoverable, and it publishes a log of what has disappeared. It was written up in Hackaday in August 2026.
This is his work, not ours. This site fetches the log weekly, credits it, and uses it for one narrow purpose: to say what a survey of live repositories structurally cannot.
Who has removed the most
Repositories deleted, by organisation
The twelve organisations the mirror has seen remove the most. Counts, not proportions.
Raw counts follow the size of the estate, so the more telling figure is the proportion. The mirror's own README puts informatics-lab at 100% deleted, moj-analytical-services at 75.6% and nhsdigital at 33.9% - against HMRC's 9.8%, which tops the raw count.
The shape of it over time
Deletions by recorded year
Mostly inferred dates - read the shape, not the years.
The 2020 and 2021 peaks are largely the mirror catching up with history rather than a burst of deleting: when the mirror started in April 2021 it had to infer when already-missing repositories vanished, and it did that from archive.org snapshots and last-commit dates. Which brings us to the caveat that matters most.
What this data cannot tell you
Only 96 of 1,566 deletion dates are exact. The rest are inferred - 67% from archive.org snapshots of the organisation or repository page, the remainder from the last git activity seen before the repository went. A repository that stopped being committed to in 2019 and was deleted in 2024 will usually be recorded against 2019. The timeline above is therefore a reasonable shape and a poor set of dates, and no individual date in it should be relied on.
Deletion is not misconduct. This needs saying plainly, because a page counting deletions by department invites the opposite reading. Most of these are dead prototypes, duplicates, repositories moved to another organisation, or things taken down for entirely correct security reasons - removing a repository is a reasonable response to finding a secret in its history, and an organisation that never deletes anything is not thereby more trustworthy. Neither this page nor the log can distinguish a tidy-up from a cover-up, and it does not try.
What the log supports is one narrow, checkable claim: this repository was public, and now it is not. That is enough to close the hole. A finding about a repository does not evaporate when the repository does, because the mirror still has it and the log still records that it existed.
What this project will do with it
The honest position is that we are not using it for much yet. The log is fetched weekly into the dataset and this page reports it. The obvious next step - checking whether the repositories an organisation removed were systematically worse than the ones it kept, which is the question that would actually justify the suspicion - requires analysing the mirrored copies rather than the log, and that is a larger piece of work than reading a JSON file.
Until then, this page is a caveat with numbers attached: every score on this site describes the estate that survived.
Data: uk-gov-mirror/deleted-repo-report by Jonty Wareing, fetched 2026-08-14. Counts are of repositories the mirror has observed disappear from a UK government organisation; organisations it does not mirror are not represented. Percentages are against the organisation's repository count as recorded by the mirror, which will differ slightly from the live count shown elsewhere on this site because they were taken at different moments.