What actually changes hands.

Someone else can own it, run it, and prove they own it.

The inventory, and the layers a technical acquirer audits — evidenced below, except those marked open. Both of those are ownership hygiene rather than engineering, named here and detailed under NDA.

1.1.0 — eleven of eleven criteria met

1.0.0 was the phase closing. That number was reserved for eleven stated exit criteria, all eleven of which are met and dated. A version that moves without a criterion moving is a version nobody can read — so it moved when the last of them did. Release 1.1.0 is a patch on that line: corrections and fixes, with no criterion reopened.

Whichever path, this is what you would have to have.

The rule corpus

One hundred and sixty-nine rules across thirty-seven policies, each citing the instrument or market convention it derives from, and twenty-four statutory values each carrying its source, its publication date and the text it was read from — present on every one. Knowing which rules the statute requires, and what each governed value is, is research. It is the component a well-funded team cannot compress, because the work is reading law rather than writing code.

The credential rail

Standards-native rather than bespoke: the wallet protocols on both sides, SD-JWT VC issued and ISO mdoc verified alongside it, and the revocation the standard specifies. Eight runs against other people's suites, on their terms — and the second format proven against a credential the European Commission's own reference issuer produced.

The verification apparatus

What makes every claim here checkable by someone who did not write it: the suites, the gates that fail a build when a rule loses its citation, and a public register anyone can re-verify without asking. It cannot be bought separately from the thing it verifies.

The reasoning

Seventy-five decision records, including the turns that were taken and then corrected. A team starting fresh pays for those twice — once to make them, once to find them again.

Twenty-two items, against published standards.

The checklist was written, committed and dated before the audit began and was not amended while it ran — anything found afterwards is disclosed rather than added to it. A checklist that grows as you read is not a stopping condition.

For a decisioning product the image is not the asset. The governed rulebook is — so a delivery standard applied only to the container secures the engine and leaves the substance untracked.

None of it has to be taken on trust.

The release pipeline verifies its own output using only the public half of the signing key, and these are the commands it runs. ⚠ They are also the commands you would run — with the public key and registry access, which are shared at the owner's discretion rather than published. The software is private; what it can be checked against is not.

The same job then feeds the verifier a rulebook altered by one byte and requires the check to fail. A check that cannot fail proves nothing, so the pipeline proves its own can.

And the room the artifact arrives in.

Delivery is half of it. The other half is the record that travels alongside — re-verified against the running system rather than re-read.

• Re-verifiedSix dated or counted claims in the dossier were re-measured against production. Five had moved — and not one was found by re-reading the sentence that stated it.
• Carried riskEvery open finding is disclosed with the reason it is carried rather than fixed. A gate fails the build when an open finding is missing from that page, or when a dated disclosure has matured and still reads as future.
• The clockMost of the governed-value attestations come due on 1 January 2027; past a sixty-day grace the issuer refuses to start. Left alone it would announce itself only in a boot log, so it is written down with its date.
• Cover sheetOne page linking what was built, what was bought and what was reused, and pointing at where each transferability layer is evidenced.
• Every advisoryA register of the known vulnerabilities in what ships, with a disposition for each — including the ones that need no action, which is the half a scanner alone never produces. It is also the machine input to the gate, so a disposition nobody wrote is a build that does not pass.

Four decisions, taken on what the industry does.

• Key-based signingNot keyless. Keyless binds a workflow identity into a public certificate log, and these repositories are private, so there is no private keyless path. It costs nothing in assurance level.
• No transparency logThe same reasoning: signature entries would put artifact hashes and a repository name into a permanent public log. Trusted time comes instead from an independent RFC 3161 authority, which the standard treats as co-equal to log inclusion and which publishes nothing. The choice is carried by a committed configuration file, so it can be read rather than inferred — and reversed with one edit by a buyer who goes public.
• A five-year floorStated with the factors that point beyond it. The commitment is the acquirer's to make, so it is written as a floor and a rationale rather than a promise on their behalf.
• The rulebook travelsFolded into the existing release steps rather than made a separate ceremony. What makes the product valuable ships under the same signature as what runs it.

Stated here rather than found later.

• Build levelSLSA Build L2 is met. Signing runs in a job that executes no build steps, so build-time code from a dependency cannot reach the key. L3 is not met, and its requirements conflict with a private estate rather than being a level not reached. L3 asks for the signing key to be held by the build platform's own service account and every field of the provenance to be generated in a trusted control plane the tenant cannot influence; ours is a key we manage, used in our own workflow, with the predicate produced by a script in our own repository. The route that would satisfy it publishes the repository name to a public transparency log, and the reference implementation refuses private repositories by default for that reason. The job isolation above is hardening, not the control the standard specifies.
• Not publicly discoverableSignatures are not published to a transparency log, so nobody outside can enumerate or monitor what this key has signed. That is the point — the estate is private — but it is also the cost: if the key were ever misused there is no public record in which to notice it. Holders of the public key can verify what they were given; they cannot audit what they were not.
• Verification is not anonymousThe images are private, as the repositories are, and the public key is shared at the owner's discretion rather than published. So verifying any of the above takes one conversation first — deliberately, and it is said here rather than discovered. What a grant buys is the ability to check every claim on this page independently; what it does not require is trusting us about any of them.