Site for EncodeBench — The Axiom Foundation's benchmark for how well AI models encode statutes into cited, executable rules.
Sister site to policybench.org: PolicyBench asks whether a model can compute the law; EncodeBench asks whether it can write it as code.
The benchmark itself (suite manifest, harness, board fold) lives in
axiom-encode — see
benchmarks/uk_model_capability_v1.yaml, the eval-suite and eval-board
commands, and docs/model-capability-eval.md. This site displays it; the
harness is the source of truth. src/data/uk-v1.ts mirrors the suite
manifest and must be updated with it.
bun install
bun run devVendored Axiom Foundation design tokens (statute paper, legal ink, warm
amber) from axiom.org/packages/ui — sync src/app/globals.css
manually when the family tokens change. Logo assets are the canonical files
from the axiom.org repo.
Vercel, axiom-foundation scope, domain encodebench.org.