fabsssss/uk-evaluation-atlas

UK Government Evaluation Evidence Atlas

An open, classified atlas of 1,770 UK government evaluation publications, harvested from the GOV.UK Search API and labelled by evaluation type and declared method. A map of what government evaluates, who commissions it, and how openly it states its methods.

Publications	1,770
Document types	research, independent_report
Year range	1996–2026
Source	GOV.UK Search API
Licence	Open Government Licence v3.0

The finding

Impact evaluations dominate the published record (886 of 1,770), ahead of process (391), mixed (139), feasibility (108) and economic (91). But the sharper result is about transparency: only 11% of publications declare a recognisable method in their title or description. Where a method is stated, qualitative (60) and survey/monitoring (50) lead; randomised designs are named in just 19. The largest commissioners are DfID, DfE and DWP.

Honest about method

Labels are assigned from each publication's title and description (its own declared framing), by a locally hosted open-weights model, not from a reading of the full report. The atlas therefore measures how evaluations present themselves in the public catalogue, which is exactly the surface a searcher or an automated evidence-synthesis tool sees first. Publications the model cannot place are labelled unclear, not guessed.

Files

• data/atlas_summary.json — counts by type, method, commissioner and year
• data/evaluation_atlas.csv — one row per publication with labels and URL
• data/classified.jsonl — full classified records
• scripts/harvest.py, scripts/classify.py, scripts/build_atlas.py

Reproduce

python3 scripts/harvest.py    # harvest from the GOV.UK Search API
python3 scripts/classify.py   # classify (local open-weights model)
python3 scripts/build_atlas.py

Contains public sector information licensed under the Open Government Licence v3.0. Independent, self-initiated open research by Tesseract Academy.