The two-witness rule

Why refuse to convict on one truthful witness when a second one adds nothing new?

Deuteronomy 19:15 lays down a rule that sounds almost arithmetical: a matter is established on the evidence of two or three witnesses, and never on one. Numbers 35:30 applies it to capital cases with no exceptions. Rome kept a version of it in the maxim testis unus, testis nullus — one witness is no witness — and the United States Constitution still carries one, requiring two witnesses to the same overt act before anyone can be convicted of treason.

But consider what the second witness actually contributes. If the first told the truth, the second describes the same event, in the same place, at the same time. The court learns nothing about the world it did not already know. So on what grounds does adding a redundant copy of a message turn an unusable statement into a sufficient one?

Start by asking what a court is uncertain about. There are two quite different uncertainties, and the answer depends entirely on which one is in play.

The first is uncertainty about the event: what happened. On that, the second witness genuinely adds close to nothing. The second is uncertainty about the channel: whether this testimony is a report of the event at all, rather than a mistake or an invention. Ask what a single witness's account looks like from the bench when it is fabricated. It looks exactly like a truthful one. That is the difficulty — not that one account is thin, but that from inside, a true single account and a false one are indistinguishable.

Now what does duplication do to that? It cannot confirm the claim; it can only compare two copies. But comparison is a real product. If the copies disagree, you have learned something you had no other way to learn, and if they agree, you have learned that any fabrication had to be manufactured twice, by two people, consistently. The second witness is not evidence about the defendant. It is evidence about the first witness.

Which tells us immediately what the rule depends on. Two copies drawn from the same source are worth exactly one. So the rule is really "two independent channels", and everything else in the surrounding law turns out to be machinery for enforcing that independence. The rabbinic courts examined witnesses separately, under a fixed interrogation about time and place — which year, which month, which day, which hour, which location — so that agreement could not be reached by improvisation in the room. Relatives of the parties and of each other were disqualified. And Deuteronomy 19:16-21 supplies the sharpest device of all: a witness proved to have been elsewhere — a zomem witness, one whose story is refuted rather than merely contradicted — receives the penalty he sought to impose. That converts perjury from a cheap gamble into the most dangerous seat in the courtroom.

Then what happens when the two copies disagree? Notice that redundancy of two can detect a fault but cannot locate it: nothing tells you which witness is wrong. The halakhic answer is not to pick the more plausible one but to discard the pair — a contradiction on an essential detail invalidates the testimony outright. That looks wasteful until you ask which kind of error the system is willing to make. Discarding good evidence costs you convictions. Choosing between two accounts on a hunch costs you the guarantee. The rule sacrifices the first to protect the second.

And it is worth being clear that the resulting rarity of conviction was noticed and, by at least some voices, welcomed. The Mishnah preserves the remark that a court putting a man to death once in seven years is called destructive, with Rabbi Elazar ben Azariah raising it to once in seventy, and Rabbi Tarfon and Rabbi Akiva saying that had they sat on such a court, no one would ever have been executed. The tradition also records disagreement with that view, which is the honest picture: the evidentiary bar was high, and there was argument about whether it should be higher still.

Three refinements. First, the rule is about witnesses, not about evidence in general: the biblical and rabbinic systems restrict it to direct eyewitness testimony to the act itself, which is why circumstantial proof, however strong, does not substitute for it. That narrowness is a design choice, and a costly one. Second, "two adds nothing" is true only if you assume the first witness is truthful — the assumption the whole rule exists to avoid making. Third, redundancy of two is a detector, not a corrector. Systems that want to identify which copy is wrong need three, and it is worth noticing that the text says "two or three" rather than "two", leaving room for the case where the third breaks a tie.

• Why circumstantial evidence was excluded even when it was overwhelming, and how later courts worked around that.
• Three-copy schemes: what changes when a system can correct rather than merely detect.
• The English and American treason clauses, and why that one offence kept the two-witness bar when others dropped it.
• Corroboration requirements in modern law and the debates over their disparate effects on which crimes get prosecuted.

Every term the collection defines is gathered in the glossary.