The previous brief was about where the observer stands. This one is about what the observer reads. An observer can only stand as far outside a system as the system's records allow, because verification works on records. When records support outside verification, outside observers become possible. This brief describes how records reach wider circles of readers, one grade at a time.
The Handelsregister
In 1861 the German states adopted a common commercial code, the Allgemeines Deutsches Handelsgesetzbuch. It established the Handelsregister, a public register of merchants. The register let a stranger confirm a firm before doing business with it: read the entry at the court, and the firm's existence and its authority to sign were established facts rather than claims. Credit could extend to firms the lender had never met, because the facts were checkable. The weaving company that lends this house its name was founded one year later.
Three Grades of Evidence
The register illustrates three grades of evidence, and each grade serves a different reader. Operator-asserted evidence serves the operator: internal logs answer internal questions, and inside a single organization that is often enough. Participant-verifiable evidence serves the parties to an interaction: each side holds enough material to confirm the shared history, so cooperation does not rest on trust alone. Evidence verifiable by strangers serves readers who were never part of the interaction: an auditor or a procurement reviewer can confirm the record from the artifact and a public anchor, without access to the operator's systems. RFC 9334 formalizes these roles for modern attestation. Article 12 of the EU AI Act asks for records of systems that cannot be replayed, because when a decision exists only at the moment of effect, the record of that moment is the evidence that remains.
The loop from the previous brief and this ladder fit together. The loop names the phases of action under challenge. The ladder describes who can read the challenge material. Each grade widens the circle of possible readers, from the operator, to the participants, to anyone.
The Second Grade
Wirken's audit log is built for the second grade. The gate commits an action and the commitment lands in an append-only log. Entries chain by SHA-256, and the chain head carries an Ed25519 signature. Anyone holding the chain and the operator's public key can confirm the history is complete and unchanged, because each entry commits to everything before it. A counterparty can verify the shared record. What the chain cannot provide by itself is standing with a stranger: someone outside the interaction has no way to place the chain in time or to know it is the version others saw. That is not a defect in the chain. It is the boundary of the grade.
The Third Grade
The third grade adds an external anchor, and the standards for the anchor were completed this year. The SCITT architecture is RFC 9943. COSE Receipts became RFC 9942 in June, and the hash envelope became RFC 9995 in July. Together they define a transparency service: an append-only public log that accepts a signed statement and returns a receipt proving inclusion. An operator registers a chain head and keeps the receipt. A stranger checks the receipt against the service's public keys and learns that the chain head existed, in that form, at registration. The hash envelope keeps records on premises: the anchor holds a digest, and the records themselves never leave the operator. The receipt travels; the data does not.
The pattern already works across organizations. At IETF 126 in Vienna, seven independent parties ran agent-action records through this substrate and reached byte-agreement across four codebases. Each party verified records the others produced, using only public artifacts. Their working notes carry a useful lesson for anyone building on this. Twice in one day a pinned artifact changed underneath its name, once as a regenerated bundle under an unchanged filename and once as a force-pushed commit under a pinned SHA. Re-hashing caught both. Verification rests on digests, not names.
Anchoring fits Wirken's design because it is evidence export, and evidence export works by mediation. Wirken extends by mediation rather than by hosting foreign code, and anchoring follows the same shape. The transparency service contributes no code and sits outside the dispatch path. If the service becomes unavailable, the gateway runs unchanged. The anchor raises the grade of the evidence and touches nothing else.
What a Receipt Says
It helps to be precise about what a receipt tells its reader. A verifiable record establishes what happened: which action, in what order, under which authorization. It does not establish that the decision behind the action was a good one. A regulated-sector consultancy recently mapped eleven UK financial-conduct obligations onto anchored agent records, and the split was clean. The factual obligations were verifiable by a supervisor from the artifacts alone, and the quality judgments remained with the supervisor. Cryptography carries facts up the ladder. Judgment stays with people. A reader who knows the difference can evaluate any evidence claim on its merits.
The Handelsregister separated the private from the public cleanly. The books stayed private, the facts a stranger needed became public, and commerce grew on that foundation. The same separation is now available to agent infrastructure.
MIT licensed.