GovBench
An author-reported multiple-choice benchmark of UK government knowledge, described as the precursor to CivBench.
ProjectSources & confidence: Limited evidence
Checked 2 October 2026 · AI-assisted review
Author-reported benchmark size, scores and motivation for the later CivBench project.
- Liam Wilkinson — I Gave an AI a Civilization to Run
Author-reported benchmark size, scores and motivation for the later CivBench project.
What this check doesn’t establish
- No dataset, evaluation scripts, public release or independently reproduced scores were located.
- Superseded describes the author’s research direction, not a verified release lifecycle.
- The author’s ORCID has not been independently verified.
Confidence applies to the scope above, not every claim made by the underlying source. How confidence works
- Kind
- benchmark (own work)
The June 2026 essay reports the dataset and scores and explains the move towards CivBench. No inspectable public evaluation dataset or reproducible release was located in this review. “Superseded” records that change in the author’s research direction.
In the essays
- I Gave an AI a Civilization to Run. It Built a Nuke.
Read the essay
People
- Liam Wilkinson
Built by
Record history
Catalogue source: knowledge/projects/govbench.md
Generated by claude-code/claude-opus-5 on 2026-07-25