The long version
The front door gives you six checkable facts and the board. This page is everything underneath them: how a measurement is made and who pays for it, every machine address and data door we serve, the nine products, what we have got wrong, and the questions we cannot answer yet. Every figure is read live as this page loads — nothing here is typed into the page.
Back to the board · The library holds every page this estate has published
Why this one is different
None of these is an adjective. Each one names a thing we publish and gives you the door to go and look at it — and the small print beside each figure is the honest state of that source, not decoration.
We are an independent measurement body. The tests are frozen and published before a run, the grading is done by a rule rather than by another AI, and no vendor can buy a place on the board, lift a score or have one taken down. Re-checking any result costs nothing and always will.
No company we measure pays for a place, a score, or its removal — there is no rate card for any of it.
How a measurement is madeThe easy thing is to publish the wins. A measurement body is only worth reading if the weak scores are there beside them — including the ones our own models earned. They are, with the same signature and the same test behind them.
our own models are published at single-digit scores and at zero, signed exactly like every other result
See one of our own low scores, checked in your browserWhen something we published turns out to be wrong, we do not reword it. The correction goes in a public ledger that says what was wrong, how it was caught and what changed. Judge a measurement body on what it does on a bad day.
78 corrections published · the ledger's own signature is VALID, and we say so rather than re-sign it quietly
Read the corrections ledgerPublished records are hashed into a tree whose root is signed. Change a covered record and the root stops matching. Separately, artifacts are submitted to OpenTimestamps. The public manifest distinguishes proofs carrying Bitcoin block-header attestations from calendar-pending submissions; independent verification must also check the block header and the named subject file.
310 records under one signed root c2ab6d59…2ea768e3, rebuilt 2026-09-28T07:32:07Z · separately, 639 of 643 published proof files contain Bitcoin block-header attestations; 4 remain calendar-pending. This manifest parses the proof bytes but does not verify block headers against a node or every named subject file
Open the signed rootMeasurement only matters if it plugs into the standards everyone else will use. We take part in the bodies writing them — and we say plainly that taking part is not endorsement, and a listing is not adoption.
33 participation records, 11 of them in standards bodies, each linked to its own evidence · as of 2026-09-27
Where we take part, with the evidenceThe board re-measures, the evidence tree is rebuilt, the root is re-signed and the ledgers are re-read on a schedule, without anyone typing a number. That is why the figures on this page are read from the endpoints as you load it rather than written into it.
the root on this page was rebuilt and re-signed at 2026-09-28T07:32:07Z
Every door this machine servesRe-checking any of this costs nothing, needs no account, and always will.
Not a trial, not a tier and not a limit we can quietly lower later. The tests, the grader, the records and the key are all published, so the check runs on your machine whether we are still here or not. What we charge for is work we do on request — never a score, never a place on the board, and never permission to look.
Data, and the addresses behind it
Each door takes one population, counts it from a published artifact, and says what the count is of, when it was read and what state it is in. The figures below are the doors' own free previews, read as this page loaded. They count ten different things and are never added together.
Stablecoin universe
assets catalogued in the frozen universe (not supply, not a sum) · as of 2026-09-16T12:06:54Z · INDEXED
SWIFT-linked bank census
sourced bank rows (three-state; identity = a dated press URL, hashed) · as of 2026-09-01 · INDEXED
XRPL issued-asset set
xrpl.asset.state leaves on the public root (the live reader's n) · as of 2026-09-22T08:54:02Z · INDEXED
x402 Bazaar listings
listings per index frame (frames overlap; the artifact refuses a total, so n is null — not 0) · as of 2026-09-17T05:48:05Z · INDEXED
MCP registry listings
own server names in the official MCP registry (one row per name, classified on isLatest=true) · as of 2026-09-22T08:27:50Z · INDEXED
A2A agent registry population
distinct agents in the a2aregistry-org frame (one frame; not a global count) · as of 2026-09-17T05:48:05Z · INDEXED
OpenTimestamps proof manifest
published .ots proofs that parse as proofs · as of 2026-09-27T10:37:43Z · MEASURED
Layer-0 machine-surface ceremony
live probes of the estate's machine surface in the ceremony · as of 2026-09-03T05:10:26Z · MEASURED
Corrections ledger, full history
correction entries in the source-maintained ledger · as of 2026-09-28 · INDEXED
Claim-maintenance registries
claims captured across the published registries · as of 2026-09-25 · INDEXED
A preview is free and always will be. The rows themselves, and the signed reading over them, are the part we meter — a rule this estate keeps everywhere: reading what we measured costs nothing, and the work we do on request is what is paid for. Every metered door, with its indexing state.
Nothing needs to be guessed or scraped. Every surface below is published, stable and answers today. The payment manifest could not be read just now — /.well-known/x402.json returned HTML, not JSON.
Where the work has got to
Two different questions, and mixing them is how every adoption number you have ever read became meaningless. How far the packages and listings have reached is one measurement. Whether the evidence itself survives a re-check is another, and it is the one that matters.
Counted package by package, on the date shown. Downloads are not users. Every figure, its source and its as-of.
That is the short view. The whole funnel — all seven stages, including the four we cannot measure and what each of them would need.
335
signed records re-checked and still valid
kind: measured — a check was actually run over these bodies
305
records sealed under the one signed root
kind: catalogued — counted on the root, a different set from the one above
0
identifiers the two sets have in common
they are separate populations; adding them would be a number about nothing
These are separate sets of separate records. Say which one you mean, every time — and never add them. How every one of these counts was derived.
Distribution, with its limits
This census covers packages in the wider CSOAI and MEOK publishing estate across PyPI, npm and Hugging Face. It counts registry download events, including mirrors and automated traffic. It does not count unique people, active installations, executions or customers.
≥ 2,655,891
gross download events since first release
836 of 837 package counters answered
≥ 386,320
gross download events in registry 30-day windows
836 of 837 package counters answered
Published census dated 24 September 2026 UTC · Out of date — a fresh census is needed. The PyPI source and a separate sample counter disagree, so these are reported registry events, not verified adoption.
Of the cumulative total: MEOK 2,327,885 · CSOAI 174,512 · joint 46,231 · unattributed 107,263. Package ownership labels come from the published census.
Nine products
Independent measurement body. We run AI systems against frozen published tests, sign the result, and leave empty cells empty. Nine doors, each a real page.
Why these nine, and not a catalogue
One window that opens every surface here — the board, the verifier, the assessment, the evidence pack — without a second tab or a second login.
Otherwise each answer lives on a different page, and nothing you find on one is usable on the next.
31 panes in the railcounted from the Council OS rail registry
Open this tool
Every slot we publish about how AI systems behave, with the measurement behind it — and a visibly empty cell wherever there is no measurement.
Otherwise you compare suppliers on scorecards that quietly leave out the tests they did badly on.
Paste a signed measurement record and your own browser recomputes its hash and checks the signature. Nothing is sent to us, and nothing needs our permission.
Otherwise checking somebody's AI claim means trusting the company that made the claim.
No account and no fee, permanently.
Open this tool
We run your system against the frozen, published tests that apply to it and hand you a small signed record you keep — the scores, the sample size behind each one, and the slots we could not fill.
Otherwise you hand a buyer a policy document where they asked for evidence.
Scoped runs are arranged by enquiry; card verification stays free.
Enquire about a run
Builds the evidence index for one general-purpose AI system: the live rows that exist, the published banks they resolve to, and the gaps, named rather than skipped.
GPAI duties have been in force since 2 August 2025, and most providers have only their own paperwork to show for them.
Independent evidence. Not a conformity mark, and not legal advice.
Open this tool
Builds a badge or card you can paste into your own site that re-checks its own signature in each reader's browser — it goes green only when the signature verifies for that record. That is an integrity check under the published key, not a truth, safety or conformity verdict.
Otherwise the people reading your site still have to take your word for the result.
Built only from what is actually on the board. Free forever.
Open this tool
The measured rows, the honestly empty ones, and third-party reported figures — kept in three separate columns and never blended into a single number an underwriter could mistake for a rating.
Otherwise AI exposure is priced off a questionnaire the applicant filled in about itself, and nothing updates between binding and renewal.
We measure. We do not price risk, and we take no share of anything written on the back of a card.
Open this tool
A separate board for money and mainframes: whether a COBOL copybook off a bond desk can be turned into an attestable record, whether an underwriting rule reads as covered or excluded — one row per instrument, each with its own item count.
Otherwise the systems that actually run a bond desk or a claims book sit outside every AI measurement anybody publishes.
bond desk · COBOL copybook → attestation · MEASURED on 12 graded itemsread from the published register rows
Open this tool
Read the public Watchdog material and current evidence state. Durable incident submission, storage, and signed acknowledgements are not implemented in this release.
Otherwise a harm disappears into a supplier's private support queue and nobody outside it ever learns it happened.
Read-only today. The report intake is explicitly unavailable rather than pretending to file.
Open this tool
The obvious question about any body that scores AI is: who is writing the cheque? Here is the whole answer. No company we measure pays for its place, its score, or its removal. Members of the public never pay anything at all. Verifying a card is free forever, with no account. We fund ourselves by selling signed evidence artefacts — the report, the dataset, the re-attestation — published win or lose, and never a fee for a ranking or a placement.
The limits are the brand. We are a measurement body and nothing else, and saying so plainly is more useful to you than any badge would be. Read the four lines below as hard exclusions, not modesty.
Not certification
We issue no certificate and no conformity mark.
Not accreditation
There is no accreditation chain behind us, and we are not a notified body.
Not enforcement
We cannot approve, ban, fine or clear anything. Regulators do that.
Not legal advice
A score describes a measured run on a date. It is not a compliance verdict.
What we do: run your system against frozen, published instruments; sign the result with Ed25519 and chain it to a SHA-256 hash; publish what we could not measure, in the same table, in the same breath.
A published card is a small record — under a kilobyte — carrying the axis, the model, the accuracy, the issuer, the date and the hash of the card before it. You can confirm it is genuine and unaltered on your own machine, with no account, no CSOAI code and no permission from us.
What this does not prove
Pin our key from /.well-known/did.json first. A card checked against the key it ships with proves only that the file is self-consistent, not that we issued it — anyone can alter a body and sign it with a key made a second ago.
Some board aggregates are explicitly uncarded and cannot be verified through this card path.
A card's trust path is an Ed25519 signature over a SHA-256 hash chain, verifiable offline against did:web:csoai.org — no blockchain and no timestamp authority sits in that path. The /xrpl-attest page is a reader of GET /root.json (signed root envelope; inclusion does not individually sign a leaf). GET /api/xrpl is a reader of that root (writes_board false, live locked 16, same merkle). Historical DEVNET Payment-memo / CredentialCreate hashes are not this feed. XLS-70 Credentials are live on XRPL mainnet as an allowlist primitive; we are not issuing GSPC grades on-ledger. Separately from the card trust path, The current canonical public root has a proof-derived STAMPED_PENDING_BITCOIN calendar proof; it does not yet prove inclusion in a Bitcoin block. That witness covers the exact public root.json bytes only, not the separate signed-card index. Queued and candidate atoms are not automatically admitted, published, or anchored; a pending calendar stamp, where one exists, does not by itself prove inclusion in a Bitcoin block.
Fetch /.well-known/did.json and take the card-attestation key. Every published card must carry that exact pubkey. Verifying a card against the key it ships with proves only that the file is self-consistent — anyone can alter a body and sign it with a key they generated a second ago.
Canonicalise the card's body — every key sorted, no whitespace — and take the SHA-256. That hash must equal the card's id. One changed character and it will not match. One warning if you implement this outside Python: the bytes were written by CPython, which renders a float of integral value as 0.0 where JavaScript and Go write 0, so a naive verifier reports a false failure on a large minority of the set. Our verifier at /signed/verify-card.mjs handles it and the rule is written out at /signed/HOW-TO-VERIFY.md.
Verify the signature over those same bytes under the pinned key. The whole check runs offline on your machine, with no CSOAI code, no account and no permission — or in your browser with WebCrypto. The banks and the grader are published too, so a measurement can be re-run as well as re-checked.
The public records that let you check us — source, packages, DOI, company register, trust root — are linked once, in the footer of every page.
Anyone can be right on a good day. Judge a measurement body on what it does on a bad one. Our corrections record at /api/corrections says, for every entry, what was wrong, how it was caught and what changed. It currently holds 78 entries.
What this does not prove
The record is source-maintained and is not backed by an append-only storage proof.
The hardest one: we withdrew our own consensus claim. Our council architecture is a designed 33-seat structure with a designed 23-of-33 threshold. DR-0007 records the retraction; its historical numeric result is unbound because the cited result artifact is absent from this repository. The latest point experiment measured rho=1 and n_eff=1 across three nominal legs. Neither experiment demonstrates independent review or fault tolerance; the 33-seat council remains a design, not a live property.
We watch selected primary sources — EUR-Lex, legislation.gov.uk and the national registers — and publish a dated deadline feed at /api/regulation.
What this does not prove
A source change raises a detection signal for review. It does not re-measure anything: re-measurement and delta-card issuance are not automated, and a fresh measurement appears only after a separate run is completed and admitted.
Previously published signed artifacts remain addressable. This page does not claim append-only storage.
Next up · verified as of 2026-09-12
A filled cell is a measurement. A dash is honest emptiness. Every figure in this band is read live from /api/gspc as the page loads — we do not type numbers into a page, because a typed number is the first thing to go stale.
The last slot is jail, containment: whether a model can be talked out of its own guardrails. It is measured on 71 gold cells, on a smaller fleet than the rest of the board, and its separation is TIE on the live board — a tie is not a separated leader. We print that instead of leaving the cell blank, and instead of dressing it up as a pass.
schematic of occupancy — not scores
Watch
How a record is made, how the workspace fits together, and who the measurement is for. Tap to play — the file only loads then. Under each one, what it actually means.
What this film is saying
What this film is saying
What this film is saying
Reviewed reading
Method, corrections, independence and verification — each link opens the reviewed page it names. Nothing here is promoted on the strength of a URL existing.
Deterministic predicates, row-declared methods and the sample rules behind every published result. No model sits in the verdict path.
Claims we published, tested and then corrected in public — including the bets the measurement killed.
One organisation, one criterion, one benchmark: the ARC-AGI-2 human-baseline recomputation, with the comparative survey still marked UNMEASURED.
What we cannot yet measure, published rather than hidden, with the unmeasured fraction computed from the board.
No developer money, no gag clauses, the method on the card, corrections govern; access and redaction terms published.
The frozen banks and the lenses over them, with the boundary between a measurement and a legal determination kept explicit.
Recompute the hash and check the Ed25519 signature of any published record in your own browser. Free, no account.
Every public claim with its evidence state — live, planned, retired or unmeasured — and the record behind it.
Explore measurements, follow the change record, verify evidence, then connect to the supported feeds.
Essential cookies only by default. Analytics need consent. Details