The notary and the seal

Consider the oldest problem in the law of ordinary life. Two people agree to something — a plot of land changes hands, a dowry is promised, a dying man says who is to have his house — and the agreement matters most precisely at the moment neither of them can be relied upon to describe it honestly. One has died. One has an interest in remembering it otherwise. The people who must now act on the agreement — heirs, buyers, judges, tax collectors — were not in the room, and the room is gone. Human memory is contestable and private paper proves nothing, because either party could have written it, or altered it, or produced it later. What was needed was a way for an act to carry its own credibility forward through time and absence, into the hands of strangers with every motive to deny it. The answer that civilization arrived at, and kept, was not to improve trust. It was to route around it.

The answer was the notary. In Roman practice the tabellio was a professional drafter of instruments who prepared documents in due form for private parties; over the later Empire the authenticated instrument acquired growing evidentiary force. The institution we would recognize took its mature shape in the medieval Italian revival of the notariate, with Bologna a center of notarial learning and practice from around the twelfth and thirteenth centuries. The notary was not a party to the act and had nothing to gain from its content. He witnessed it, reduced it to a fixed form of words, dated it, and — crucially — entered it into his protocol, a register he was bound to keep and preserve. The instrument he issued was authenticated: first by the seal, later by the notarial sign and subscription, marks that a stranger could recognize and that were costly to counterfeit. What the notary produced was not a better promise. It was an act made into a thing that could travel.

The disinterested witness

Understand what the notary actually solved, because it is subtler than record-keeping. Anyone can keep a record. The difficulty is that a record kept by an interested party is worth exactly nothing to the party it is used against — it is the opponent's own account, and no one is bound to accept an opponent's account of a disputed act. The notary's genius was to interpose someone whose credibility did not depend on the outcome. He would not gain if the sale stood or fell; he would not gain if the will named the son or the stranger. His only stake was in his own office, which depended on his being reliable across thousands of acts he had no interest in. That is why his word could do what the parties' own words could not. A disinterested third party who witnesses and records converts a private claim into a public fact — not because he is more honest than the parties, but because he is differently situated, outside the quarrel he is asked to fix.

The protocol register did the second half of the work. A single sealed sheet can be lost, suppressed, or quietly replaced; a bound and preserved register, kept in sequence and dated, cannot be edited after the fact without the edit showing. When a dispute arose years later, the instrument in a party's hand could be checked against the notary's own retained entry. The act was thus recorded twice, by an indifferent hand, in a form where tampering left a trace. The seal and the sign made each instrument recognizable as authentic; the register made the whole body of acts tamper-evident and time-fixed. Together they produced something a private letter never could: a record whose authenticity did not rest on trusting the person holding it.

The notary's seal did not make an act true. It made it impossible to later pretend the act had not happened, or had happened otherwise.

Authenticity that travels

The deepest property of the notarial instrument is that it carries its own proof. A well-made instrument answers, on its face and from its register, three questions that a stranger would otherwise have to take on faith: when did this act occur, who made it, and what exactly was said. It answers them in a form checkable by someone who trusts neither party and was not present — which is the only form that is worth anything, because the moment authenticity depends on trusting one of the disputants, it has stopped being authenticity and become allegiance. This is the exact difference between a notarized act and a self-attested one. Self-attestation is the interested party vouching for itself: this is what happened, take my word. It is precisely the thing the notary was invented to replace, because it fails at the only moment it is needed, the moment someone disputes it.

Notice that none of this required the notary to certify that the act was wise, or that the parties were telling the truth about their intentions. He certified something narrower and more useful: that this act, in these words, was made by these parties on this date in his presence. Authenticity is not veracity. The notary made no claim about whether the promise would be kept or the sale was fair. He fixed the fact of the act so firmly that the argument could move on to whether the act was sound — instead of remaining stuck, forever, on whether it had happened at all. That is the whole economy of the institution: it takes the question of provenance off the table so that the questions that actually matter can be reached.

Notarizing a decision

Now turn the lens. A machine decision has exactly the structure the notary was built for. An act occurred — a system decided something, on some basis, under some rule, at some instant — and it will need to be believed later by people who were not present and who may have every motive to dispute it: the person denied the loan, the applicant screened out, the regulator, the court. And most decision systems are, precisely, un-notarized. They are self-attested: the system that made the decision is the same party that describes it afterward. They are preserved by the interested party, in logs that the interested party controls and can edit. They are undated in any tamper-evident sense — a timestamp the owner can rewrite is not a date, it is a suggestion. And they are authenticated by nothing that a skeptic who trusts neither party could check. This is the un-notarized paper of the pre-modern world, reproduced at scale: the interested party's own account, offered as if the offering settled anything.

What the notary supplies, and what a machine decision needs, is third-party, tamper-evident, time-fixed authentication of the act of deciding. A record of the decision that fixes when it was made, on what evidence, under which rule as that rule then stood, and does so in a form where later alteration is detectable and where the authenticity does not depend on trusting the deciding system — this is the missing witness. A Decision Receipt is, in the plainest historical sense, an attempt to notarize an act of judgment: to give it provenance, tamper-evidence, and a fixed time, so that its authenticity travels with it to parties who were not there. The point is not that the decision was right. The notary never claimed the sale was fair. The point is that its having-happened, in this form, on this date, on this basis, can no longer be quietly denied or rewritten. That is the difference between a decision you can contest on the merits and one you cannot even get a stranger to agree occurred.

We are about to make an unimaginable number of consequential decisions by machine, and to preserve almost all of them the way the pre-modern world preserved a private letter: in the hands of the party with the most reason to shade them. The notary is two thousand years old and the reason he was invented has not changed. When an act must be believed by people who were not present and cannot afford to simply trust you, the answer is not a more eloquent account. It is a disinterested witness, an authenticated instrument, and a register that will not lie about the date.

— Dispatches · Summit Cognitive

---