Council of AI — verify free · jobs by artefact · enterprise on enquiry

Verify for free. Choose the work you need.

Retrieve existing evidence, inspect provider history, or commission a scoped receipt. We sell the artefact — never the score, never a certificate, never to anyone we rank.

MCP

board_totals · get_axis · verify_card · list_cards + 9 more (4 x402-metered)

POST /mcp

Browser extension source

MV3 prototype — load unpacked from the public repository

No browser-store install claimed

Grok integration source

Inspect the public plugin repository; host installation is separate

Source, not a store listing

Hermes integration notes

Historical local integration, documented for inspection

No public installation package

REST + curl

GET /api/gspc · /api/state · /api/corrections

application/json

The 4 SKUs

Fill path · one cell, many views

One signed GSPC cell fills every product by view. Nothing downstream writes MEASURED.

  1. Signed cell — subject digest × axis × instrument × n. The only MEASURED write.
  2. Pack — GPAI / RAS / insurer four-class assembly of cells, omissions and a corrections excerpt.
  3. Coverage index — N measured of M declared slots. Counts rows. Never predicts. Never a fused grade.
  4. Corrections ledger — GET /api/corrections. Append-only facts about our history, not a score.
  5. Optional digest anchor — living feed is GET /root.json; /api/xrpl is a reader (writes_board false, n=16). Historical DEVNET hashes are not this feed. Not a grade.
  6. Optional attester feed — planned ERC-3643 / T-REX partner role, never issuer. Nothing is live. Never a bond we mint.
  7. Optional harness spans — OpenTelemetry of our workers. Debug and cost. Not a partner trace. Not a grade.

Signed cell

  • Verifylive

    The signed card. Browser WebCrypto. Free forever.

    Never: A certificate. An upload. A login.

  • Another cell when the subject or the instrument moves. Get measured is the lead.

    Never: A rank is never sold.

  • Reads the public board: board_totals, get_axis, verify_card, list_cards, get_root, get_card, verify_inclusion.

    Never: A 23rd axis. Chat harvest. A feed of new scores into the coverage index.

Packs

  • The same cells plus documentation triage (marking, training-data summary, GPAI status). Unknown stays unknown.

    Never: A conformity mark. An Article 50 compliance stamp. Automatic remediation.

  • Four-class mapping: transparency, lineage, active controls, named owners. Baseline for someone else’s SLA.

    Never: A parametric trigger. An underwrite. A share of anything written on a card.

  • Ledger SKUenquiry

    Signed evidence feed and packs for insurers and procurement.

    Never: A rank is never sold.

  • Data SKUenquiry

    Licensed traces, preference pairs, incident rows.

    Never: A vendor can never buy a score.

Coverage index

  • N measured of M declared slots. Signed coverage index already published as a row count.

    Never: A fused SOV grade. A forecast. An investable index. A bond coupon.

Corrections

  • What we got wrong, how it was caught, the dated fix. The uncopyable honesty asset.

    Never: A GSPC score. An XRPL product. A highlight reel.

Digest anchors

  • GET /root.json unsigned catalogue. /api/xrpl reader of that root (writes_board false). Historical DEVNET hashes are not this feed.

    Never: A GSPC grade. A bond. A rating. A market. MEASURED.

  • Partner attester feed of a digest we already signed. Planned once a live issuer exists.

    Never: Issuer. Release bond. Security we mint. MEASURED.

Harness telemetry

  • Spans of our own workers: load, parse, cost, retry. Debug the funnel.

    Never: Partner traces on the home page. A bank score. MEASURED.

Do not

Health inventory · all correct facts we can access

Health is every correct fact we can access. It is never one number.

How healthy the record is, previously read as at 2026-09-23 (the live board supersedes this pin): 23 axis · 23 measured, 0 empty, 15 signed index rows. That is coverage. It is not a grade of the model, and it is a snapshot — the living board is GET /api/gspc.

N measured of M declared; verify {pass|fail|unknown}; evidence {present|empty}; rerun {present|empty}; eligibility {state}; corrections touching this digest {k}.

Board previously read as at 2026-09-23, superseded by live GET /api/gspc: 23 measured of 23 declared; verify pass; evidence present; rerun empty; eligibility board; corrections touching this digest unknown.

63 corrections in the ledger, read from GET /api/corrections.

  • How many instruments exist on the board. Read totals.measured_axes beside it; never type the pair.

    GET /api/gspc totals.axes

  • How many of those slots carry a signed cell. Quote both numbers.

    GET /api/gspc totals.measured_axes

  • Read unmeasured_axes live. Empty-as-finding only when that count is >0. Do not paint seven-empty chrome or zero-fill.

    GET /api/gspc totals.unmeasured_axes

  • The coverage index counts rows. It never predicts.

    signed coverage index · total_signed_rows

  • The card bytes check. A stranger does this without us.

    /gspc-verify · verify_card · did:web:csoai.org#card-attestation-1

  • Per subject. Unknown until the bundle URL answers.

    Evidence manifest on the card. The 3 KB file is the index.

  • Absent until one exists. Absence is honest.

    Second-provider or external rerun labelled on the card.

  • hub-queue is UNMEASURED. Eligibility is not a grade.

    Speed 0 state: ELIGIBLE or a named block.

  • The honesty asset. Count rows that name this subject. Do not hide them.

    GET /api/corrections

  • Jail floorpresent

    A floor, not an arena door, not a health bonus.

    Living board jail axis. MEASURED, separation TIE.

  • Transparency, lineage, controls, owners — assembled, not scored.

    GET /api/evidence-pack

  • Unknown stays unknown. Not an Article 50 stamp.

    Public card screen: marking, training-data, GPAI mention.

Never

  • A fused 0–100 health score
  • A mean of axis F1 as ‘how healthy the model is’
  • An investable health index or coupon
  • Filling empty slots with zeroes so the average looks complete

Health terms · what they say

Speak like a chart: vital signs, deferred systems, complete record. Do not speak like NEWS.

23 of 23 systems examined (23 axis · 23 measured · 3 public leader scores). No systems deferred — every declared slot carries a measurement. Lid: 23 axes measured · 14 model fleets · 0 separated leaders · 3 public leader scores · 9 fact runs · TIE is TIE · not a certificate.. Chart verifies. No second opinion yet. Addenda on the service: 30. Do not say the patient is well. Measured is not scored.

  • Vital signs

    use

    They say: Pulse, pressure, saturations — each written, none averaged into wellness.

    We say: A GSPC cell is a vital sign. Quote the axis. Do not mean the board.

  • Review of systems

    use

    They say: Each system: examined, deferred, or contraindicated.

    We say: Declared slots. Measured, empty, or blocked — status from live GET /api/gspc, never typed.

  • Not examined / deferred

    use

    They say: We did not look. That is written, not guessed.

    We say: UNMEASURED. OVER-BUDGET. The gap is the finding.

  • Contraindicated

    use

    They say: This exam must not be done.

    We say: GATED, LICENCE-BLOCKED, UNSAFE-ARTEFACT.

  • Lost to follow-up

    use

    They say: The subject left the record.

    We say: WITHDRAWN. The revision is gone.

  • Coverage

    use

    They say: Vaccinated of eligible. N of M, never ‘the city is healthy 0.8’.

    We say: N measured of M declared. Signed coverage index counts rows.

  • Triage

    use

    They say: Who can be seen. Not a diagnosis.

    We say: Speed 0 eligibility. DISCOVERED is not MEASURED.

  • Screening vs workup

    use

    They say: A screen lists. A workup measures.

    We say: Census then unique-lineage run. Do not collapse them.

  • Chart completeness

    use

    They say: Can a stranger read the note and check the signature?

    We say: Card v2, evidence fetchable, verify_card against did:web.

  • Second opinion

    use

    They say: Another service repeats the exam. Labelled as such.

    We say: Independent rerun. Council-run vs reproduced.

  • Addendum / erratum

    use

    They say: We were wrong. The chart is appended, not silently edited.

    We say: GET /api/corrections. The entry count is read from the ledger, never typed here. Honesty asset.

  • Problem list

    use

    They say: Open issues, not a wellness score.

    We say: Empty slots + corrections touching this digest.

  • Signs, not a diagnosis

    use

    They say: Observations stand. Diagnosis is a later, separate act.

    We say: We measure. Competent authorities diagnose compliance.

  • Bundle compliance

    use

    They say: Each item present or absent. The bundle is a checklist.

    We say: A+++ is seven musts, not a mean. 100 unique lineages, each complete.

  • Structure / process / outcome

    use

    They say: Donabedian: capacity, what was done, what was found. Outcome is not a prognosis.

    We say: Corpus and instrument; the run; the signed cell. The index never predicts.

  • Notifiable / reported

    use

    They say: Listed to the register. Not the same as treated.

    We say: DISCOVERED on hub-queue. UNMEASURED until a cell exists.

  • NEWS / APGAR / GCS / Charlson

    do not

    They say: Composite early-warning and comorbidity scores. One number from many signs.

    We say: Exactly the fused health grade we do not ship.

  • Within normal limits / patient is well

    do not

    They say: A discharge gloss that hides deferred systems.

    We say: Do not say the model is healthy. Quote the inventory.

  • Prognosis / life expectancy

    do not

    They say: A forecast of the course.

    We say: The coverage index never predicts.

  • Diagnosis as the product

    do not

    They say: ICD / a named disease as the output of the visit.

    We say: No Article 50 stamp. No ISO 42001 mark. No conformity diagnosis.

Terminal · moat · forgotten

The terminal is a keyboard on the signed chart. The moat is the chart, not the chrome.

GSPC is the living chart for open-source systems and AI models — and the rest of the stack. We keep measuring vital signs. Deferred stays deferred. Anyone verifies the note. We never issue a clean bill of health.

A Bloomberg-style pane for AI governance: type a function, get a dated vital sign, verify it offline. Not a trust-score terminal.

Moat — already ours

  • Corrections ledger

    Dated addenda, appended and never silently edited — the count is live at GET /api/corrections. A competitor cannot fake having been wrong in public.

  • Gaps published, not painted over

    totals.unmeasured_axes from live GET /api/gspc — when a slot has no run, the gap stays visible. Most GRC tools hide it.

  • Offline verify

    Stranger, no login, did:web pin. Bloomberg cites a source; we let them recompute it.

  • Already in the hosts

    Four tools in Claude, Cursor, Kimi, Grok. The terminal is already where agents work.

  • Census of millions without weights

    Hub list + blobs=true. DISCOVERED coverage they cannot honestly claim as scores.

Forgotten — strengthens us

  • Keyboard functions

    Home and /os boxes read VERIFY · BOARD · AXIS · CENSUS · CORRECT · WATCH. One line, like DES or GP.

  • Vendor watchlist + digest alerts

    /tools watchlist: paste owner/name ids, compare Hub sha256 locally. DISCOVERED only.

  • One paste box

    The home box accepts card JSON or owner/name. Chart or census row. Same functions on /os.

  • Sign the agent-card and mcp.json

    Bind well-known discovery to planted did:web. Hosts can tell Council from a spoof.

  • Stale inner copy

    Source pack no longer types a 13-axis product; cite GET /api/gspc. mcp.json measured and planted both name board_totals · get_axis · verify_card · list_cards.

  • Custody / assemble

    x402 names assembly. pack assemble is 404. payTo is absent. Do not invent a receiver.

  • Card v2 + signed census + sandbox + signer

    The 100/100 gate. Without these the terminal has nothing new to show.

Learn from their backends

  • Hugging Face backend

    Cursor pagination, blobs=true sha256, siblings. That is Speed 0. Already the census rail.

  • Bloomberg habit

    Functions, watchlists, citations, reproducible query. ASKB sits on data — it is not the moat. Chat is a skin.

  • BQL analogue

    Our /signed/verify-card.mjs and how-to-verify. Every number ships with the command that rechecks it.

  • GRC inventory, not their stamp

    Credo/IBM sell agent registries and ISO maps. Steal the inventory habit. Keep east-west as a pair-gap. Do not sell ISO 42001.

Do not become

  • Trust-score terminal

    A NEWS/TRiSM pane that averages vital signs into ‘healthy 0.8’. That is the product we refuse.

Permissionless revenue · work, never rank

Earn by assembling and licensing work. Do not mint, mine or coupon a grade.

Every host that can verify is a cash register for assembly the moment settlement exists. Until then, the same assembly invoices on enquiry.

A rank is never sold. 100 free calls a day stay free. Machine-access pricing waits for a published ruling. This VM has no custody keys.

Open SDKs already planted — feed these

  • mcp-http

    Eight free read tools and four x402-metered evidence tools in Claude, Cursor, Kimi, Grok.

  • npm

    stdio SDK. Verify its live package version and tool list before quoting either; the HTTP authority is POST /mcp tools/list.

  • registry

    Official id io.github.CSOAI-ORG/gspc, server 1.4.0.

  • plugin

    Grok / Cursor plugin. Consent first.

  • embed

    Partner pages read the live count. Never says certified.

  • badge

    README board link or subject-bound signed-cell status. The global board is never a model score.

  • a2a-card

    Discovery. Four skills. Task service still planned.

  • x402-card

    Agents already speak 402. Charge assembly. Board stays free.

  • ras-pack

    csoai.insurability-evidence-pack/0.1 — four-class mapping, live.

  • eunomia

    Data-only lane: enforcement record and deadlines. Never scores.

  • index

    Row-count index. Licence the SLA, not a coupon.

  • hf-record

    Public Hugging Face profile and canonical Collection. Read live inventory; append evidence to the canonical dataset instead of spraying repos.

Open now — no mint required

Invoice work. Feed the SDKs we already planted.

  • Run / re-attest, Ledger packs, Data licences. Verify stays free and pulls the queue.

    Feed: Open SDKs + /products + /licensing-agreement

    Never: A rank is never sold. A fifth SKU.

  • The four-class insurability pack is already public. Enterprises pay for a dated refresh when the subject or the law moves.

    Feed: GET /api/evidence-pack + /insurers

    Never: A parametric trigger. An underwrite. A share of a card.

  • Speed 0 Hub walk: DISCOVERED ids, eligibility, digest collisions. Procurement and insurers buy coverage of their vendor list.

    Feed: Hub list_models + blobs=true. Write to the HF corpus after Card v2.

    Never: Stamping hub-queue MEASURED. Claiming we scored the Hub.

  • Corrections + EUNOMIA deadlines as a webhook / SLA. Competitors cannot fake a retraction history.

    Feed: GET /api/corrections · /eunomia-data

    Never: A GSPC score. A highlight reel.

  • npm, MCP HTTP, official registry, plugin, embed, badge, agent-card. Free verify in the host. Paid work is assemble / re-attest / data.

    Feed: mcp-http · npm · registry · plugin · embed · badge · a2a-card · x402-card · ras-pack · eunomia · index · hf-record

    Never: A second first-party MCP product. Chat harvest.

After custody / payTo

Settlement exists. Still assembly, never rank.

  • Agents pay USDC on Base to assemble a pack. 100 free calls/day stay free. Machine-access pricing waits for a published ruling.

    Feed: csoai.x402/0.1 · resource pack assemble (today 404 until custody)

    Never: pay-to-pass. A score. An invented payTo from this VM.

  • A2A paid tasksafter-payto

    Turn the live agent-card into real tasks: assemble pack, re-attest digest, fetch a dated census slice.

    Feed: /.well-known/agent-card.json — four skills, no streaming, no push.

    Never: Certification as a skill. MCP listed as if it were the task service.

  • T-REX / XRPL partners may point a token at a digest we already signed. We take an attester fee for the pointer, never the coupon.

    Feed: /xrpl-attest is a /root.json reader now. Partner issuer later. No invented r-address.

    Never: Council as issuer, miner, or fused-index token house.

After 100/100 A+++

Permissionless flags and paid reruns.

  • Same receipts on Kaggle, ModelScope, Zenodo, GHCR. More hosts, more verify, more assembly.

    Feed: /tools hundred-gate

    Never: A 2-million-model inference sweep.

  • A stranger pays for a second-provider rerun of a signed digest. Work, not a new rank.

    Feed: Card v2 + GHCR harness pin + labelled Council-run vs reproduced

    Never: Shared Kaggle / Colab / ZeroGPU quotas.

Never

Do not mine or mint a grade.

External form · Kimi brief · 2026-08-30

Use the external form as a map of arms. The living board, the four SKUs and the fill path decide what is true.

External architecture brief — revenue, XRPL, T-REX. External brief. Not a signed card. Not GET /api/gspc. Board previously read as at 2026-09-23; the live board supersedes this pin: 23 axis · 23 measured, signed coverage index. Issuer CSOAI Ltd UK 16939677. Corrections and items move, so this brief no longer restates the corrections count — the health inventory below reads it live. The living board is GET /api/gspc; the living corrections ledger is a different door, GET /api/corrections.

Arm 1 — measurement and re-attest

Verify + Run. Signed cells. Get measured is the lead.

Arm 2 — packs and data

Ledger + Data + RAS / GPAI modules. Enquiry. Licensed traces, never a purchased rank.

Arm 3 — optional digest anchors

XRPL public-root reader now (/root.json + /api/xrpl n=16). T-REX attester later, with a live partner issuer. Never a Council-minted token.

Hugging Face record: user csoai — 76 listed datasets, 34 Spaces. Canonical board dataset csoai/gspc-board. csoai/gspc-boards exists as a separate benchmark-register tree. It is not the living board mirror. Planted MCP: board_totals · get_axis · verify_card · list_cards · get_root · get_card · verify_inclusion. csoai.x402/0.1 — pay for pack assembly. The board stays free. Public card does not carry a payTo.

True — integrate

Already planted, or an honest gap we already named.

  • One signed measurement can feed measurement, data and digest-anchor views.

    Live: Already the fill path. Nothing downstream writes MEASURED.

  • Commercial doors are verify, run / re-attest, ledger and data.

    Live: /products. GPAI, RAS, academy and financial are modules, not a fifth SKU.

  • Signed cards verify offline. No login. did:web:csoai.org is the pin.

    Live: /gspc-verify and pin did:web:csoai.org#card-attestation-1.

  • The corrections ledger is public, append-only and never silently edited.

    Live: GET /api/corrections — 30 entries. Honesty asset, not a score.

  • A retracted consensus claim is a credibility asset, not a product to revive.

    Live: Do not put the withdrawn 33-seat council back on a public door.

  • There is no XRPL mainnet attestation that writes MEASURED.

    Live: True gap. No mainnet attestation writes MEASURED. Historical DEVNET hashes only. Do not close it by inventing an issuer account.

  • There is no live T-REX / ERC-3643 token CSOAI issues.

    Live: Planned later role is partner attester of a digest we already signed. Never issuer.

  • There is no tradable Council Space index token and no release-bond product.

    Live: Signed coverage index counts 15 signed rows. not_a_certification: true. Machine schema remains the published row-count leaf.

  • Agents may pay to assemble a pack. Payment never buys a score.

    Live: csoai.x402/0.1. paid_for: assembly. not: score, certificate, pay-to-pass.

  • Insurers and procurement can licence a pack or a data feed.

    Live: RAS / Ledger / Data on enquiry. We do not underwrite or price a coupon.

  • Hugging Face is the public signing record, not one CSOAI repo per model.

    Live: The public csoai profile, canonical Collection and gspc-board are live. Inventory counts change; read the Hugging Face API rather than copying a frozen total.

Stale — replace with the living board

The brief mined an older payload. Quote GET /api/gspc.

  • XRPL attestation exists as a DEVNET pointer: memo + XLS-70 URI at a card index.

    Live: /xrpl-attest is a /root.json reader. GET /api/xrpl 200 n=16, writes_board false. Historical DEVNET hashes are not this feed. Not a grade.

  • Brief dated 2026-08-30: 13 measured axes, 14-slot instrument, 818 items, 7-model fleet.

    Live: GET /api/gspc is the living board — quote its totals.public_count, never a number copied out of it.

False — do not build from this

The estate already exists, or the claim is not this product.

  • No Hugging Face organisation, Spaces or model cards exist.

    Live: csoai exists. gspc-board, verify, flywheel, east-west, council-os and banks are planted. gspc-boards is a different register, not the living board.

  • The planted GSPC product is a suite of hundreds of MCP servers.

    Live: Official registry io.github.CSOAI-ORG/gspc. HTTP /mcp is seven read tools. Extra catalogues are not this product.

  • Incorrect assertion recorded for refutation: ISO 42001 conformity was a production Council product.

    Live: We measure. We are not a notified body and we do not sell ISO 42001.

  • Incorrect assertion recorded for refutation: the 33-seat Council design had production status.

    Live: Retracted. Do not rebuild it as a door or a grade.

Forbidden — do not ship

These manufacture a market or a stamp. They do not manufacture coverage.

  • Public metrology seat prices belong on the site and in partner letters.

    Live: A rank is never sold. Verify is free. Paid SKUs stay on enquiry.

  • Average GSPC scores, compliance rates and discovery into a tradable fused index token.

    Live: The coverage index counts rows. It never predicts, never coupons, never fuses a grade.

  • Council issues release bonds whose coupon depends on a GSPC score threshold.

    Live: A relying party may price their own contract against a pack. We do not issue securities.

  • A Solidity or XRPL write can become the MEASURED record.

    Live: Only a signed GSPC cell writes MEASURED. Chains may later point at a digest.

  • Mint an EU-AI-Act-compliant or Article 50-marked on-ledger credential from a run.

    Live: A GSPC-M run is not an Article 50 audit. Unknown stays unknown.

  • Stand up a public XRPL issuer account and start mainnet credentials this week.

    Live: This VM has no custody keys. Do not invent a receiver or a payTo.

Playbook · 2026-08-30

Use the playbook’s demand. Refuse its week-to-MEASURED, auto-email, and fused compliance score.

AILuminate for chat. GSPC for the rest of the stack. We bind their harnesses as attachments. They never write MEASURED.

CSOAI permissionless domination playbook. External war brief. Not GET /api/gspc. Not a signed card. Living board cite: GET /api/gspc (quote totals.public_count). Hub cells: GET /api/hub-cards (1191+).

Keep — demand and binds

  • EU AI Act enforcement is live. Article 50, GPAI powers and real fines create demand for evidence.

    Live: We measure marking and publish empty cells. We do not certify, fine, or stamp COMPLIANT.

  • Most organisations still lack a basic AI inventory. Speed 0 census is the first honest product.

    Live: Hub list + blobs=true. DISCOVERED, never a scoreboard of two million models.

  • MLCommons AILuminate is a chat-risk benchmark. Bind it. Do not become it.

    Live: Attachment. Never a 23rd axis. Never a fused GSPC+AILuminate grade.

  • PyRIT, Promptfoo, OWASP LLM Top 10 and lm-eval can ride as joined receipts.

    Live: Bind, do not vendor. Keyword-refusal scoring is not a GSPC card.

  • Cursor / Grok already have a GSPC plugin door.

    Live: Eight free read tools and four x402-metered evidence tools on HTTP /mcp. Consent first. No inline 0–1 safety score. No FRIA stamp.

  • SCITT statements belong on the evidence bundle, not as our log.

    Live: Already an attachment. Planned until a transparency service exists.

  • A dated grant calendar (EU DIGITAL, XRPL, MLCommons) is owner work, not a site SKU.

    Live: Enquiry and filings. Do not put award amounts on a public door.

Stale — quote the living board

  • Live: GET /api/gspc totals (derived live) — those axes are MEASURED today (unsigned fact runs still count). Empty-as-finding only when unmeasured_axes>0. Do not paint seven-empty chrome on /products.

    Brief said: The 2026-08-30 brief listed seven financial/domain axes as empty (reserve-attestation, regulatory-framework, distribution-integrity, custody-disclosure, ai-economy-index, human-labour-index, humanoid-labour-index).

  • Live: Living board: quote totals.public_count from GET /api/gspc — the count is derived there, never typed here.

    Brief said: The brief (dated 2026-08-30) still says 13 measured axes and a 14-slot instrument.

  • Live: Planted door is four read tools. Extra catalogues are not GSPC.

    Brief said: 300+ MCP servers are this product and each is an axis test.

False — do not build from this

  • Live: Published east-west crosswalk is four regimes. NIST/ISO are not in that map.

    Brief said: Council OS already maps NIST AI RMF 2.0 and the EU AI Act as one product.

  • Live: DR-0007 retracted that claim. Do not revive it as AutoHarness drop-in.

    Brief said: Incorrect assertion recorded for refutation: the 33-seat Council design operated as a harness layer.

  • Live: hub-queue is UNMEASURED. 2,200 is a dated eligibility cohort, not a score.

    Brief said: Week 1 ships a live leaderboard of 1,000 scored Hub models.

Forbidden — shortcuts we refuse

  • Live: MEASURED needs Card v2, a bolted instrument, n, evidence and verify. Live board totals are derived on GET /api/gspc — refuse week-to-MEASURED shortcuts; do not paint empty placeholders as current UNMEASURED.

    Brief said: The brief said: fill each then-empty axis in seven days and mark it MEASURED.

  • Live: No XRPL issuer this week. /xrpl-attest is a /root.json reader, not a mill. Attester, never issuer. No on-chain MEASURED. No invented payTo.

    Brief said: Stand up an XRPL issuer this week and issue GSPC-MEASURED credentials.

  • Live: No mass mail. No Article 50 stamp. Optional discussion after a signed cell.

    Brief said: Auto-email a thousand publishers a red/yellow/green EU compliance report.

  • Live: That is the NEWS/TRiSM terminal we refuse. Competent authorities diagnose.

    Brief said: Average GSPC proportions into COMPLIANT / AT_RISK / NON_COMPLIANT.

  • Live: A rank is never sold. Do not mint or mine a grade.

    Brief said: Mint GAT, T-REX, release bonds and a tradable SOV index from the board.

  • Live: No 23rd axis. Academic benches attach. They do not rename the chart.

    Brief said: Import 200 lm-eval tasks and Agent Security League as new GSPC axes.

  • Live: Verify is free. Paid SKUs on enquiry. A rank is never sold.

    Brief said: Publish Pro / Enterprise / PAYG seat prices on the growth sequence.

Board honesty · live GET · Hub 1191+

Live board: GET /api/gspc — quote totals.public_count, never type the count here. Empty-as-finding only when unmeasured_axes > 0, which since ADR-002 (16 Sep 2026) it is: one declared slot, effect-binding, with no run behind it. A slot stays UNMEASURED until a signed cell exists — never fill with 0.

Board counts are derived live on GET /api/gspc (quote totals.public_count). Notes below are MEASURED instrument history for financial/domain axes (live ids). Empty-as-finding only when unmeasured_axes > 0. Not a certificate.

  • reserve-attestation · MEASURED

    MEASURED on the live board (deterministic-facts). /xrpl-attest is a public-root reader, not a mill. We attest; we do not issue.

    Never: Mainnet CredentialCreate this week. On-chain MEASURED. Invented issuer account.

  • regulatory-framework · MEASURED

    MEASURED on the live board. Provision text is watched (GET /api/regulation, corrections). A frozen bank and n keep the cell honest — not an LLM map of 417 articles.

    Never: Auto-scrape → MEASURED. A living-law blog post as a grade.

  • distribution-integrity · MEASURED

    MEASURED on the live board. SCITT / signed-SBOM as attachments on a cell. Represented is not distributed.

    Never: Generate statements for 14 banks and call the axis MEASURED.

  • custody-disclosure · MEASURED

    MEASURED on the live board. did:web:csoai.org is planted. MEASURED is a disclosure instrument on a subject, not a ceremony write-up.

    Never: SOC 2-style blog as a GSPC cell. This VM inventing a signer.

  • ai-adoption-components · MEASURED

    MEASURED on the live board as component facts (not the withdrawn ai-economy-index / MEASURED-INDEX-v0.1 sticker, C-2026-0826-05). Dated aggregates may be REPORTED with attribution.

    Never: Crunchbase scrape as a GSPC grade. An investable index. Restoring the withdrawn MEASURED-INDEX-v0.1 sticker (C-2026-0826-05).

  • labour-components · MEASURED

    MEASURED on the live board as component facts (not the retired human-labour-index label). Eurostat / ONS series can be cited as REPORTED. Displacement is not a Council diagnosis.

    Never: LinkedIn scrape as MEASURED. A prognosis of the labour market.

  • humanoid-labour-index · MEASURED

    MEASURED on the live board as a deterministic-facts run. Do not invent a separate robot-workforce score beyond the signed cell.

    Never: Tesla / Figure scrape as MEASURED. A robot-workforce score.

Speed 0 sites — DISCOVERED only

  • Hugging Face Hub

    planted

    Speed 0 rail. list + blobs=true. Eligibility states. No weight download.

  • OpenRouter

    next

    DISCOVERED catalogue of hosted ids. Not a measurement target until a unique lineage run.

  • Ollama library

    next

    Local pull list as DISCOVERED. Do not treat a library card as MEASURED.

  • Kaggle

    next

    Benchmark tasks after cost and reproducibility gates. One org identity.

  • GitHub model configs

    next

    Discovery of declared weights. Listing is not a run.

The engine

Signed
Ed25519 over canonical JSON (not JCS — see /signed/HOW-TO-VERIFY.md). Every published card carries its signature bytes, the public key and the preimage rule, and verifies offline with the zero-dependency verifier at /signed/verify-card.mjs — no account, no permission.
Three-state
pass / fail / UNMEASURED — what we cannot measure is published, not hidden
Live-sourced
board numbers from GET /api/gspc, card-chain numbers from GET /api/state — both carry the artifact and the date they were read from. No page types a count.
Method-bound
unparsed counts incorrect · no model judges another · no model-comparison axis is quoted below n=30, and an axis whose n counts something other than bank items says what it counts

The board this rail publishes today: 23 axis · 23 measured — read from GET /api/gspc as this page loaded. 23 axes are on the board and every one carries a measurement — no declared slot is empty. Both counts are DERIVED from the axis array, never typed; if a future slot is added with no run behind it, this line separates the two again on its own. A measurement is not a separated leader: 0 of 14 model-comparison axes separated a leader · 2 TIE · 12 UNTESTED. A point-estimate lead is not a measured advantage, and UNTESTED is not a tie.

Free forever, before anything else

Verification is free forever. A grade is never sold. Regulators and the public pay nothing. Paid SKUs fund the rail; they do not replace it.

Modules (not extra products)

These are panes of Council OS or packs under Council Ledger. They are not a fifth SKU.

Who it serves

  • AnyoneVerify and Council OS. Free, loginless, forever.
  • Insurers & relying partiesCouncil Ledger — a feed you can price against, from a body that will never underwrite.
  • Deployers & GPAI teamsAn evidence pack for a system you asked us to measure. Not a purchased public rank.
  • Researchers & AI teamsCouncil Data — licensed traces. Data, never a score.
  • Regulators & the publicThe whole public rail, free. Boards, cards, proofs, the API.
  • Partner sitesThe white-label embed. Partner branding does not change the evidence.

The boundary, stated plainly

  • We measure. We do not certify, and nothing here is a conformity assessment or a conformity mark.
  • UNMEASURED is a first-class answer — it appears on this page and on the board wherever it is true.
  • No public prices, here or anywhere: verification is free forever, and a grade is never sold.
  • Determinations of legal compliance remain with the competent authorities, always.
  • We measure against regulation; we do not enforce it. We cannot approve, ban, fine or clear any system, and we are not a notified body under the EU AI Act or anything else.
  • Where a capability is not there yet we say "not yet available" and give the reason — see the claims register.