We do not certify. We measure. A reference index of every AI framework, charter, regulation, and benchmark — and a measured, signed, contamination-aware evaluation layer over it.
The EU AI Act GPAI obligations started 2 Aug 2026. Regulators and enterprises now need verifiable evaluation evidence — and the model industry is being caught gaming benchmarks. A measured, contamination-aware evaluation layer is the direct answer.
| Asset | What |
|---|---|
| Reference index | 662 sourced instruments (frameworks, charters, regulations, articles, sectors, benchmarks) over MCP/A2A |
| Measured gauge | 30252 live records, 97 signed cards, 18 axes (Fisher-Rao-style trust gauge) |
| Arena Elo | Bradley-Terry on our own measured pillars — not market Elo, not crowd-sourced |
| Contamination register | 36 benchmarks tracked (12 designed-resistant, 5 high-leak) — the anti-Goodhart guard |
Benchmarks are being gamed. Our contamination register + designed-resistant probes + signed runs are the honest counter. We report measured, never certified — unmeasured axes stay UNMEASURED, and no score is inflated by contamination.
Compliance evidence for EU AI Act GPAI, a trust-conformal 90/10 router, a measured arena, and data licensing of consented eval-transcripts. Measurement, not certification.