The witness and the oath

Before a court could weigh a witness's account, it had to solve a prior and harder problem: why believe the account at all. Words are cheap. Anyone can walk into a room and say what happened, and saying it costs the speaker nothing whether the words are true or false. A tribunal that simply accepted whatever it was told would be a machine for laundering convenient stories. So the law reached, very early and across almost every legal tradition, for a device that looks primitive and is in fact rather deep. It made the witness swear. The oath was not decoration. It was the mechanism by which a person's testimony stopped being free.

The swearing bound the witness's word to a consequence. In its oldest religious form the consequence was divine — the oath called down judgment for a lie — but the enduring, secular achievement was to attach an earthly penalty as well. To testify falsely under oath became a punishable offense. English law developed the crime of perjury over centuries, hardening from ecclesiastical and customary sanction into a settled common-law and statutory offense by the early modern period, and every mature legal system carries some version of it now. The particular statutes and their numbers matter less than the shape they all share: a person who takes the oath and then lies has committed a distinct wrong, prosecutable in its own right, apart from whatever the lie was about.

What that accomplishes is best understood not as moral theater but as economics. An oath plus a real penalty converts testimony from cheap talk into a costly signal. Before the oath, a true statement and a false one cost the speaker exactly the same — nothing — so the words carried no information about which they were. After the oath, a false statement carries an expected cost that a true one does not. The witness who lies is now spending something: exposure to punishment, forfeiture of standing, the risk of prosecution. Because honesty is cheaper for the witness than deceit, the tribunal can rationally extend a credence it could never extend to unsworn words. The oath does not read the witness's mind. It changes the witness's incentives, and it does so in a way the court can see.

An unsworn word costs nothing whether it is true or false, which is exactly why it tells you nothing. The oath makes the lie expensive, so the truth becomes worth believing.

Making a word cost something

Notice what the oath is not. It is not a test of accuracy and it does not verify the content of what is said. A sworn witness can still be honestly mistaken, and a determined liar can still perjure himself and hope not to be caught. If we imagined the oath as a truth detector we would find it a poor one. But that was never its function. Its function is to put a price on the lie and, just as importantly, to fix that price to a named person. The oath is taken by someone in particular, identified, present, answerable. It ties this specific statement to this specific individual, who now has skin in the game. Testimony, having been made costly, becomes testimony someone is standing behind.

That combination — a penalty for falsehood and a named party bound to it — is what lets a decision rest on testimony at all. Strip either half and the structure collapses. A penalty attached to no one deters nobody. A named speaker who risks nothing by lying is back to cheap talk. It is only when a particular, identified person has staked something real on the truth of a particular statement that the statement acquires the property we loosely call reliability. Reliability here is not a fact about the witness's character. It is a fact about the arrangement: interest has been aligned with honesty tightly enough that the account can be relied upon.

This is worth separating cleanly from a nearby institution it is often confused with. A notary does something different — a notary authenticates that an act occurred, that a signature is genuine, that a document is what it purports to be. That is authentication of an act. The oath is not about authenticating an act; it is about binding testimony — a claim about what happened — to a penalty and to a named party. The witness is not certifying a document. The witness is putting a personal cost behind an assertion, so that the assertion can be weighed as something more than words.

Not a guarantee of truth, but a price on the lie

It helps to be honest about the limits, because the honesty is the point. The oath guarantees nothing about the truth of any given statement. What it guarantees is that lying is no longer free — that the witness's interest has been bent toward honesty and that there is a named person the system can hold to account if the account proves false. That is a modest claim and a durable one. It has survived for millennia precisely because it does not overreach. It does not pretend to see inside anyone. It arranges the incentives so that, on average and over time, sworn testimony is worth more than unsworn testimony, and it makes visible who is answerable when it is not.

The staked, named witness is the load-bearing element. Everything downstream in a trial — cross-examination, impeachment, the penalties for contempt — assumes there is a specific person whose credibility can be attacked and who bears a consequence for the attack succeeding. Accountability needs an address. You cannot hold a rumor to account, because a rumor is no one's. You can hold a sworn witness to account because the sworn witness is someone, and that someone has already agreed, by taking the oath, that a false word will cost them.

An oath for the machine's account

Now consider the accounts that machines give of themselves. A system reports what it did. A log attests to its own operation. A vendor assures a customer that the model behaved as specified. These are all, by default, cheap talk in the strict sense the oath was invented to cure. Nothing is bound to a penalty. No named party stakes anything on the truth of the report. The account can be produced freely whether or not it is accurate, and a false account costs its producer exactly what a true one does — which is to say nothing — so the account, on its own, tells you nothing about which it is. A self-attested log is not testimony under oath. It is an unsworn statement by a party with every incentive to be reassuring.

The oath is the model for what such an account is missing, and the model is precise. A decision record worth relying on has to do what the oath did: bind the account to a consequence and to a named, accountable party, and make the stake real rather than rhetorical. That the record be tamper-evident is not a technical nicety here; it is what makes the stake genuine. If the account can be quietly revised after the fact, then no one has truly staked anything on the version you are reading — the penalty attaches to a statement that can be swapped out. Tamper-evidence is how you fix a particular account to a particular party so that they cannot later disown it. It is the machine-age equivalent of the witness having sworn to this statement, in front of the court, on the record.

The distinction that follows is the one worth carrying away. There is an account someone answers for, and there is an account no one staked anything on, and they can look identical on the page. The first is a record a named party has bound their accountability to, tamper-evident so the binding holds — the thing a Decision Receipt is meant to be, and the reason its standing does not rest on the reader trusting the system's good nature. The second is a self-description offered under no obligation to be true, disownable the moment it becomes inconvenient. The law learned, a very long time ago, that testimony you can rely on is testimony someone has been made to pay for lying about. We are relearning it now for machines, and the temptation is the same one the oath was raised against: to accept a fluent account because it is offered confidently, from a source we are asked to trust, with no one bound to its truth and nothing at stake if it is false.

— Dispatches · Summit Cognitive

---