Code · HDRUK · UK Public Sector GitHub Observatory

Language figures come from GitHub's own per-repository byte counts, which measure source rather than a single headline label. File counts and repository sizes come from the mirrors themselves. The two tell different stories, and the difference is the point.

The codebase

languages in use

250.5 MB of source

62.9k

files at HEAD

across every analysed repository

0 B

on disk as mirrors

full history, every branch and tag

4.4 GB

largest repository

papers

What stands out

One archived repository holds 35% of all the bytes.

papers is 4.4 GB of git - a 2013 static snapshot of www.gov.uk, not source code. It single-handedly makes HTML the largest language by volume (36% of all source bytes). Any size or language statistic that does not exclude it is measuring an archive, not a codebase.

Sizes here are GitHub's reported figures.

This organisation was harvested with blobless partial clones, which fetch every commit and file path but not file contents, so there is no local size to measure against. Checked elsewhere in this dataset against 1,526 fully mirrored repositories, GitHub's reported size is accurate to within about six percent.

Ruby is the house language, but its share is falling.

Ruby is the primary language of 0% of archived repositories but only 0% of live ones. Python, Go, TypeScript and HCL have taken the difference - the signature of a shift from monolithic Rails publishing apps toward data pipelines, infrastructure-as-code and platform tooling.

Most repositories are small; a handful are enormous.

49% of repositories are under a megabyte of git data. The estate is not a monolith - it is a very long tail of small services, gems and prototypes around a few large publishing applications.

Languages, two ways

By volume of source

Bytes of code. Dominated by archived static HTML.

By number of repositories

How many repositories name each language as their primary one. A better guide to what engineers actually work in.

Live estate versus archive

The same language ranking, split by whether the repository is still active. The shift is the clearest signal of technical direction in the whole dataset.

Active repositories

Archived repositories

Scale

Repository size distribution

Git data per repository, log-ish buckets.

Commits against files

Each circle is a repository, sized by number of distinct authors. Both axes are logarithmic. The diagonal band is the normal relationship between codebase size and churn; outliers above it are long-lived apps with heavy iteration, below it are dumps and imports.

The shape of a repository

Directory layout is the cheapest reliable signal of what a repository actually is. app/ with config/ and spec/ is a Rails application; src/ with test/ is Java or Node. Counted by how many repositories contain each top-level directory at HEAD.

Most common top-level directories

Most common file extensions

By total file count across the estate.

The twenty largest repositories

Repository	Language	Git size	Commits	Files	Status
papers	Python	4.4 GB	4,022	52	active
datasets	Python	3.7 GB	1,622	52	active
manuscript_code	R	628.3 MB	15	13,164	active
mvcm-api	Python	384.3 MB	257	3,457	active
synthea	Java	309.8 MB	5,362	1,552	active
gatk	Java	263.3 MB	14,548	1,427	active
openpathology-web	Jupyter Notebook	181.3 MB	168	1,798	active
EPACTS	C++	181.0 MB	106	518	active
RDMP	C#	157.2 MB	9,479	4,676	active
ringbp	R	136.6 MB	19	33	active
covid19_asymptomatic_trans	R	129.1 MB	92	19	active
openprescribing	Python	127.9 MB	7,040	958	active
HDRUK_AppliedML	Python	112.2 MB	41	8	active
software_differences_paper	Jupyter Notebook	109.2 MB	27	73	active
data-profiles	Python	97.8 MB	7	19	active
hie-training	Jupyter Notebook	91.5 MB	3	76	active
Biomarker_Identification	-	90.3 MB	4	11	active
researchfish-data-analysis	HTML	76.0 MB	136	133	active
ensembl-vep	Perl	71.3 MB	1,979	465	active
synthea-profiling	HTML	69.9 MB	23	135	active