We measure whether AI systems follow specific legal provisions — 417 statutory provisions crosswalked across the EU, UK and US — on published harnesses, and we sign every result so it can be verified independently.
No. It is a measurement on a published harness with a stated method and confidence interval. Accredited certification bodies can consume our signed records as audit evidence — we are the measurement layer, not the certifier.
A signed record that stays current. When a regulation or model changes, corpus-watch flags the affected evidence and re-measurement is triggered. A PDF in a drawer cannot do that.
The measurement globe plots signed measurements where they happened — the same data as the reports, made navigable.
Benchmarks, harnesses and raw results are public on Hugging Face, Kaggle, and archived at Zenodo.