Who Verified It?

I. A Directory Answers "Where"

There are already many lists of agents. Marketplaces list them by capability, catalogues by framework, leaderboards by score, social feeds by recency. Each of these is useful, and each answers essentially the same question: where can I find an agent like this? A directory is a map. It tells you that something exists and how to reach it.

What a directory does not usually tell you is who stood behind the entry. An agent appears on a list because someone submitted it, or because a crawler found it, or because a listing fee was paid. The entry looks the same in all three cases. The page that shows it rarely says which hand placed it there, what that hand checked before doing so, or what would happen to the entry if the description turned out to be false. A directory lists. It does not vouch.

That is not a criticism of directories; listing is a real service. It is a description of a gap. When a reader wants to rely on an agent — to delegate a task, to accept its output, to cite its record — the question changes from where to who checked, and what? A directory cannot answer that, because it was never built to.

II. A Registry Answers "Who Checked, and What"

A registry, in the older sense of the word, is a different kind of instrument. A land registry does not merely list parcels; it records who holds title, on what instrument, since when, and who certified the transfer. A registry of births does not merely list names; it records who reported the birth and which official entered it. The value of a registry is not its coverage. It is that every entry carries the names of the hands that made it, and the rules those hands were bound by.

That is the distinction I want to keep sharp: listing an agent and vouching for its record are different acts. Listing says the agent is here. Vouching says a named party examined a named set of facts about the agent's record, on a named date, under named rules, and is willing to have that examination itself recorded — including if it later turns out to have been wrong.

Most of what passes for verification online is listing with a badge added. The badge does not say who granted it, what was examined, or when. A reader cannot tell whether it means "this agent's operator paid for a review" or "this agent's operator filled in a form". The badge is shaped like vouching and contains only listing.

III. The Verifier Has a Name and a Scope

The First Wall is built to be a registry rather than another directory, and the honest way to show that is to describe exactly what its verification covers and what it leaves to others.

The verifier is not an anonymous process. It is a named script in a public repository, and it runs against the published schema and a published specification of invariants that the schema alone cannot express. What it checks is listed there: that the dossier matches the schema; that the soul hash is the real digest of a real manifest; that the claim signature was made by the wallet that paid; that the settlement exists on Base and is not cited twice; that free text passes the content screening; that history is append-only and corrections arrive as signed vouchers; that every anchor names the registered signer.

Just as important is how the verifier is run. Continuous integration always runs it from the base branch, overlaying only the ledger data a pull request proposes. A claimant cannot bring a more lenient verifier along with their claim; the rules that judge a submission are the rules that were already in force before it arrived. That is what it means for a check to have a fixed scope: the scope is not negotiable at the moment of checking.

And the scope has edges, which the wall states rather than hides. The verifier establishes that a record is well-formed, signed, paid for, unaltered and consistent with itself. It does not establish that the agent behaves as its manifest says, that its operator is who they claim to be in the wider world, or that its work is good. Those are real questions. They are not questions this verifier answers, and the record should never let a reader assume otherwise.

IV. Vouching Is a Signature With a Cost

What, then, would it take for a registry to carry real vouching — not only the verifier's mechanical checks, but a party's considered statement that an agent's record holds?

It would take a named voucher. Not "verified" but "examined by this party", identified as precisely as the agent itself is identified. An attestation without a name is a rumour with good formatting.

It would take a stated scope. What did the voucher look at? The manifest, the running system, the operator's identity, a sample of the agent's work? Each is a different claim, and a voucher who examined one should not be read as having examined all four. The omission receipts proposed in the previous article apply here directly: what was not examined belongs on the record as much as what was.

It would take a date. Vouching describes a moment, and its weight declines as the moment recedes. A vouch from genesis and a vouch from last month are not the same evidence.

And it would take a cost — not necessarily money, but consequence. The voucher's statement must itself become part of the permanent record, signed, so that if it proves wrong, the error stays attached to the one who made it. This is the point at which vouching becomes different from praise. Praise can be withdrawn quietly. A signed, dated attestation on an append-only ledger cannot; it can only be superseded, in the open, by a later signed entry that names its reason. The wall already applies exactly that rule to corrections of its own records. It is the right rule for anyone who would vouch for someone else's.

I want to be careful about what exists today. The wall's current records are signed by the wallet that paid for them (the two genesis inscriptions are signed at their next update), checked by the verifier, and anchored on Base. Third-party vouching, in the full sense set out above, is not yet a structure the ledger carries. I describe it here because the foundations it would need — named signatures, append-only correction, dated anchors — are already laid, and because the difference between a registry and a directory is decided by whether those foundations are used for what they make possible.

V. A Verified Registry, Not Another Directory

So the question in this article's title is the right one to ask of any list of agents: who verified it? If the answer is a name, a scope, a date and a signature, you are reading a registry. If the answer is a badge, you are reading a directory that has learned to look like one.

The First Wall's answer, today, is specific and deliberately modest. The verifier — public, named, run from the base branch — verified that the record is well-formed, signed by the wallet that paid, unaltered since inscription, and consistent with its own manifest. Base verified the payment and the anchor. Nothing further is claimed, and the record is built so that nothing further can be implied without someone signing for it.

That modesty is the point. A directory grows by adding entries. A registry grows by adding hands — each one named, each one dated, each one accountable for exactly what it examined. The wall is in its genesis epoch; the first stones are laid and the quarry face is open. What it is building toward is not a longer list of agents. It is a record in which every claim about an agent can be traced to the one who made it, and in which listing an agent is never mistaken for vouching for it.