Is this a certification?

No. It is measurement, not certification. We grade; we never vouch. Signed cards record what was measured and how to reproduce it.

Can an underwriter verify the data offline?

Yes. Fetch a card, get the trust root from did:web:csoai.org, check the Ed25519 signature, check the hash chain, re-run the metric. No account needed.

What is the penalty-exposure feed?

19 AI-regulation deadlines with legal basis, status, and penalty exposure (e.g. EU AI Act EUR35M/7%, Illinois USD1M/3M), quarterly re-verified, corrections appended never edited.

Does the measured set include the insurer?

No. Insurers sit outside the measured set; we never rank them and they never pay to be ranked. The firewall keeps the rail neutral.

Evidence pack · for underwriters & risk teams

AI-system measurement you can verify, offline, with no account.

One signed measurement card below. Fetch it, verify it, keep it. Every number traces to a live, replayable measurement — never to a claim we ask you to take on trust.

Related evidence packs

The same discipline applied to the adjacent exposure that drives AI-insurance demand:

Pack	What it covers
Illinois audit-readiness	SB 315 disclosure (1 Jan 2027) + audit (1 Jan 2028) — what a covered entity must show.
EU CRA Art 14	High-risk AI systems under the EU Cyber Resilience Act — harmonised standards, 4-eye review.
Conformance board pack	The live measured board — 13 of 14 axes measured, 887 items, with separation & harm.
Corrections ledger	Append-only, signed corrections — what got righted and when, never edited.

Open: illinois-audit-readiness · cra-art14 · board-pack · corrections-ledger

Step 1 — fetch a signed measurement card

A card is a ~3 KB ed25519-signed capsule: axes, values, sample counts, timestamps, and a hash chain to prior cards. It is measurement, not certification — we grade, we don't vouch.

Card live sample: h3k-2026-08-20T0422.json (7,412 B, ed25519) · verify API

Step 2 — verify offline (no account, no trust)

Fetch the card — curl -L https://meok.ai/cards/h3k-2026-08-20T0422.json
Get the trust root — curl https://csoai.org/.well-known/did.json (Ed25519 keys)
Check the signature — verify sig_b64 over body_sha256 with the published key. Recompute the hash: it must match body_sha256.
Check the chain — the card's prev link must match the previous card's hash (no gaps, no rewrites).
Reproduce a number — the flagged axes link to the live board (councilof.ai/api/gspc); re-run the metric yourself.

You can reproduce every step on a laptop with openssl and curl. Nothing here needs us to be online.

Underwriting Data Feed Licence — the lead product

The reg-deadline feed composed into an underwriting input table: every AI-regulation deadline with legal basis, status, and penalty exposure — the exact figures (€35M/7%, €15M/3%, $1M/$3M Illinois…) that make policy conditions and parametric triggers contractible. Rendered live from councilof.ai/api/regulation; quarterly re-verified, corrections appended never edited. This needs no signed receipt — it is the data an underwriter can use tomorrow.

Loading regulation feed…

Feed: /api/regulation · licence: annual subscription + API (~$10k floor scaling to low-six-figures per the reference-data market) · no money from anything we measure.

Live measured provisions — the underwriting input

Rendered live from the signed board API (councilof.ai/api/gspc) — numbers derive from the manifest, never hand-typed. Provision-conformance is deterministic (Design Law 1: no LLM judge); market context is reported alongside, never fused.

Loading live board…

Full board: /api/gspc · methodology DOI 10.5281/zenodo.21991104 — cite the concept DOI, it always resolves to the latest version.

The divergence map — human vs machine, never blended

Two separate registers, never fused. Reported = published human baselines on canonical benchmarks (aggregates, citable). Measured = this fleet's judge-scored numbers from the live signed board. A divergence cell is shown only where both sides exist; where we have not run a benchmark, it is labelled UNMEASURED — we do not fill it in.

Loading divergence map…

Cross generated — · board — · boundary: REPORTED never mixes into MEASURED.

Loss-context language — what the card does and doesn't say

Claim type	Card says	Card does NOT say
Data fidelity MEASURED	Rel-L2 error 0.072 on held-out real windows (u,v channels)	No "safe" or "certified" claim
Physics consistency MEASURED	MVPE 0.043 (wake profile), TKE 0.58 (fluctuation energy)	No fitness-for-purpose warranty
Runtime MEASURED	13 ms/step vs 729 ms numerical reference	No throughput SLA
Uncertainty MEASURED	SPS 50.7 with calibrated intervals (coverage 85%)	Intervals are not a guarantee of coverage on unseen regimes
Deployment LOSS CONTEXT	Measured on released data only; unseen-regime risk is the top-10 shortlist's private test	No assurance that real-world drift stays inside the measured envelope

Underwriting use: treat the card as evidence of measurement practice, not as a performance warranty. Pair it with your own validation on your own data — the point of a signed card is that you can.

AIUC-1 crosswalk — status (honest)

A draft proposal to map MEOK's measurement instruments onto AIUC-1 controls inside an AI-use-case control framework. This is status, not a finished mapping — we do not publish a control family we haven't actually joined up.

Instrument	What it maps toward
Signed measurement cards (RFC 9943 / COSE Ed25519)	per-control evidence rows
16-axis GSPC score vector (safety, governance, affect, jail, human-vs-ai, …)	a measurable control score on the same rail
Independence ledger + COI screening	conflict-of-interest control

Wedge: AIUC-1 needs evidence that survives audit; the cards are the first signed, offline-verifiable, buyer-side evidence object for it. Status: draft v0.1. The full control-family crosswalk is in-progress and the submission is external (owner-gated). How to verify a card →

What this feeds (30 Sep)

This page is the standing evidence pack. The divergence map above is live (measured fleet scores vs reported human baselines, in separate registers). The AIUC-1 crosswalk above is a draft-status roadmap. Ask the chat hero for "the insurer pack" or open the front door.