Verifiability
w 0.40receipts passing all checks ÷ receipts sampled × 100
Every check — digest recomputes, Ed25519 signature valid, chain intact, issuer key resolves in the registry — is recomputed by the scorer, so it can't be self-reported.
Adoption leaderboard
Get listed — and prove you earned it.
Every score is recomputed from your own receipts. You can't buy a place, inflate a number, or self-report your way up — you can only evidence it.
Four numbers. All recomputable from receipts.
Coverage
w 0.35receipts that verify ÷ decisions in declared scope × 100
The denominator is your declared scope — public and spot-checked.
Human oversight
w 0.25receipts with valid human_review ÷ high-risk receipts × 100
Maps to EU AI Act Art 14 — sealed inside the digest, so it can't be added after the fact.
Overall = 0.40 × Verifiability + 0.35 × Coverage + 0.25 × Human oversight.
Enroll in five steps — and model your score now.
Registry listing is a governed grant — currently a pilot — not an automatic result of sealing receipts. Sealing locally is free and needs none of this; publishing and anchoring to the registry is the independent Seal layer.
Register your issuer key
Publish your org and public key to the registry so receipts can be attributed.
Declare your scope
Name the decision type, jurisdiction and expected volume. This sets the Coverage denominator — public and spot-checked.
Publish a receipt feed
Expose a public feed of your receipts so anyone — including the scorer — can pull and recompute them.
Run the conformance check
The public scorer samples your receipts, recomputes digests, resolves keys and scores all four metrics.
Get listed
A score and tier are assigned, a badge is issued, and you appear on the leaderboard — re-scored continuously.
Your live score
0—
Modeled client-side from the published weights. The real score is recomputed by the scorer from your receipts — recompute, not self-report.
Ready to be measured on what you can prove?
Charter pilots enroll first. You can't buy a place, only evidence one.