Peace on credit: what nobody checked about the scapegoat

Vitaly Reznik

Some ideas appeal to the powerful. This one comes from René Girard, and it is admired by people like Peter Thiel, who studied with him at Stanford and has often credited his influence. It is attractive in its darkness, and it sounds like a discovery about "how things really work".

Here it is, fairly, without caricature.

We do not desire with our own desire; we want what another wants. Hence rivalry, and rivalry is contagious: the more people want one thing, the more everyone wants it. The crowd heats up to "all against all", and order is about to break. Then the discharge comes: the hostility that was spread over everyone suddenly converges on one person. All against one. That one is declared the culprit — the source of the plague, the quarrel, the disorder — and killed or driven out. And then: quiet. The rage is spent, the crowd is one, peace is back. The corpse that a minute ago was a monster becomes sacred, because it gave peace. Out of this murder, Girard says, grow ritual, religion, culture (Violence and the Sacred, 1972).

Ugly? Yes. But "productive": violence is spent on one person, and order is bought with that one for everyone else. That is Girard's account. What follows from here is my reading of it: that this economy is attractive to power, because it explains order without conscience and presents it as a necessity of nature.

I do not dispute the description. The mechanism is real; people do this, from the schoolyard to the newspaper campaign. I want to look at one step inside it.

In the mechanism, nobody checks whether he is guilty.

Girard knew this. His answer was that the victim is innocent and the crowd does not see it. I want to say something narrower, which does not need to know whether the victim is innocent: his guilt was not established — and the mechanism runs on treating "not established" as "established".

One might object that the mechanism never claims to establish anything: its work is social, not cognitive — it restores unity, and whether the victim is guilty is beside the point. Girard himself answers this, and the answer is the hinge of everything below. His word for it is méconnaissance, misrecognition: the mechanism works only as long as the persecutors sincerely believe in the guilt of their victim. A crowd that knew it had picked someone at random would discharge nothing; it would see a murder. The claim of guilt is not a by-product of the unity — it is what the unity is made of. That is why the question "was it established?" is not beside the point. It is the one question the mechanism cannot survive.

On this page some claims are marked computed. Each is a section of a program, dilemmas/girard.py in the open repository inventor1975/ZTL, that runs the Zero-Trust Logic kernel on a model stated in the program and stops with an error if the kernel answers otherwise. The kernel is a logic with a third value, Z, for "not established", and one rule: it does not stamp a verdict that has no ground under it. A computed claim is a fact about the model, not an observation of any crowd: it shows what follows, and what does not, once the situation is written down in a stated way. Everything else on this page is my reading, and the machine has no say on it.

The crisis has no culprit

Look at the crisis of all against all. It is symmetric: nobody in it is guilty by nature. Each is a rival to each, all on equal terms. I wrote it down this way: person i is the culprit exactly when nobody else is — the same relation for every person, and nothing outside the crowd that grounds anyone's guilt. That is a model of Girard's description of the crisis at its peak. It leaves out how the crisis builds up — the contagion, the escalation — and keeps only where it ends: everyone a possible culprit, no one marked by a fact.

Then I asked the kernel who the culprit is. It abstained. For crowds of 2 to 6 people it leaves every "he is the culprit" at Z, and its passport for the crisis reads: underdetermined — as many ways to settle it as there are people; refusal until someone stipulates. In this model no culprit is forced. (Computed: section 1.) The same instrument can name a culprit: give one person a case grounded in evidence, and it does, at once. It does not here because nothing in the model grounds one. (Computed: the control in section 1.)

To get one culprit you have to choose him. With four people (any size behaves the same): stipulate any one of the four and the whole system settles — him guilty, the rest innocent — and the four settled worlds are the same world with the names swapped. Nothing in the model prefers any of them. (Computed: section 2.) Girard adds that real crowds do not pick at random: the victim carries "victimary signs" — the foreigner, the disabled, the odd one out. That decides who is picked. It does not make him guilty. As far as guilt goes, the choice is arbitrary.

Unanimity is not a check

Now look at what covers the pointing finger: unanimity. "Everyone agrees it was him." But a crowd's agreement is not a check, and we know it from small things: silence passed off as consent; "who is not with us is against us", which files the undecided among the enemies. (On an unestablished "with us", the kernel's "against us" is not established either: not-Z is Z. Computed: section 4.)

Write the accusation down honestly. Each accuser did accuse — that is a fact. And in the mimetic picture each took the same step, "his accusation makes him guilty", from the others: that is what imitation means here. It is Girard's premise — mimetic unanimity is contagion, not agreement reached separately — and a premise of the model, not something the kernel found; the claim below is conditional on it. With one accuser or with fifty the kernel gives the same answer: open, and it names the one weak link, the copied step. The count of accusers never reaches it. (Computed: section 3.) Nor does any cleverer way of combining copied accusations: across 267 random combinations in which the verdict depends on the copied step, the kernel never once says earned. At most it says true on credit — and names the step the credit rests on. (Computed: section 3, any composition.) That the same formula reads earned once the step is stamped "true" is no discovery — it is the point: the only thing standing between "guilty" and "not established" is a stamp. Accusers who each bring their own evidence are a different case: there one checked piece is enough, and the kernel says earned. (Computed: the control in section 3.) Copied unanimity is worthless not because it is many voices but because it is one voice, repeated.

That is the core of the mechanism — not even the killing, but the substitution: the unestablished is passed off as the settled. Peace is bought not with guilt — none was established — but with the appearance of guilt, assigned by unanimity minus one. That is not a payment. It is a loan.

A loan whose principal is never repaid

A loan of this kind has a sign by which you always know it: the payments keep coming due. The crisis comes back, and a new scapegoat is needed, and another — the eternal return of the victim. Girard saw the repetition clearly and explained it by rivalry regenerating; in his last books he saw the escalation running to the end (Battling to the End). The reading here is close to that late Girard: the sacrifice postpones, it does not settle. My reading is that the sacrifice cannot stop it, because it never repays the principal: the guilt it spent was never established. What keeps coming due is the interest.

The model shows why the sacrifice does not stop the return. Remove the victim from a crowd of five, and the four who remain are exactly as undecided as before: the same symmetric passport, now with four ways to settle it. The sacrifice changed nothing in the relations that made the crisis; it added no ground, so the loop it rode is still there. Ground guilt in evidence instead — each person's guilt resting on a checked fact about that person — and the system settles by itself, and removing the culprit leaves nothing undecided. (Computed: section 6.) This is a property of the model, not a history of real crises; what it shows is that, in it, a sacrifice and a settlement are different things.

And the principal can stop being repayable at all. While the victim is alive, his guilt is an open question with a way to close it: a trial could still establish it. Once he is gone, that trial is gone. If nothing else can establish his guilt afterwards, the kernel's stamp changes from redeemable to permanent: no future earns the peace that was bought with it. (Computed: section 5 — on that premise, which is mine: a later finding sometimes does clear the dead, and then it puts back an act that the model removed. Even then it clears the name; it does not repay the peace.)

The envy of the strong

My reading, and it goes beyond Girard: perhaps the debt has a deeper source. We are used to thinking that it is the weak who imitate — the one who has nothing of his own and reaches for another's. But there is a more dangerous imitation: the envy of the strong. Not "he has it and I do not", but "I would have done it better" — and here is the substitution. Not every "I would have done it better" is envy; it can be a sober estimate of one's own powers. It becomes envy when it is taken as a right — to accuse, to punish, to be paid. Then "I would have done it" is not a deed; it is a supposition taken for a fact. The weak borrow what belongs to another; the strong borrow their own unperformed greatness and demand payment on it as if it already existed. Envy is also a loan, only the borrower is sure he is the creditor.

What the kernel adds is only the form, and it agrees with it: "I am owed, because I would have done better" is open, and since the deed was never performed there is no act that could ever establish it — the stamp is permanent. (Computed: section 8, on that premise.) This envy is more dangerous than the other kind because the strong have the power to execute a sentence on a bill they have never presented — and that, by its nature, cannot be presented.

Why I want to save Girard, not break him

Because he himself was against the sacrifice. Central to his work from Things Hidden Since the Foundation of the World (1978) to I See Satan Fall Like Lightning (1999) is the claim that the Gospel revelation for the first time took the victim's side, showed the victim innocent, and by doing so exposed the mechanism and took away its hidden power. The mechanism works only while the crowd believes that the culprit really is guilty. For those who see that the guilt was assigned, the spell breaks, and the murder stops giving sacred peace: it is seen for what it is, the killing of someone whose guilt was never established.

So whoever takes Girard as an instrument — a manual for producing order through a victim — stands on the side he spent his life exposing. That is my reading of such a use, not a claim about any named person. To save Girard is to return his own blade to him and turn it against that use.

And the machine I began with is a small piece of his revelation, written in a language in which an unestablished verdict cannot pass for an established one. It does not judge history and does not know who was really guilty; it can do one thing — refuse to sign a verdict that has no ground under it. Ask it about a guilt nobody established, and it will say neither "yes" nor "no". It will say: not established — and refuse to stamp. That refusal is not weakness; it is the whole point. The mechanism lives on unanimity, and it loses its cover with the first person who refuses to stamp along with everyone else — which is also why that person is so often the next victim.

The way out

Is there a way out of the crisis other than the victim? In Battling to the End (2007), one of his last books, Girard was dark about it. A fair question to anyone who refuses the sacrifice is: if nobody pays, who pays? If the mechanism does a job, refusing it does not abolish the job; it hands the bill to someone else.

The answer is already in the model, and it is not "abstain". It is establish. Look again at the contrast in section 6: when guilt rests on a checked fact about a person, the system settles by itself, and removing the culprit leaves nothing owed. That is what a court is for. Girard saw it himself: in Violence and the Sacred the judicial system is the heir of sacrifice — the institution that ends the cycle of vengeance by having the last word. The difference the kernel insists on is in the kind of word. A trial is an act over the question of guilt; convict only on what it establishes. Evidence found: earned, a conviction. Evidence against: refuted, an acquittal. Nothing established: open — no conviction, and no certificate of innocence either; the old Scottish verdict for it is "not proven". And while the accused can still be tried, the open question is redeemable, not a debt carried forever. (Computed: section 9 — the kernel computes the rule "convict only on what is established", not how any court works; whether real courts keep that rule is a question about institutions, not about logic.) So the bill is not passed to someone else. Where proof is had, the proof pays, not a person — and what is paid with proof is not owed again.

Where proof cannot be had, what remains is the uncomfortable part: abstain. Do not choose a scapegoat. Leave the case open — endure the crisis instead of quenching it with someone else's blood. Sometimes waiting costs lives, and something must be done at once. But the need to decide is not the need to name a culprit: a dangerous person can be stopped, restrained, the others protected, without declaring him guilty of what is not established. The kernel agrees on one precise point, inside the model. In the crowd of four, no single victim is necessary for a settled outcome — for each one there is another who would do. And the unsettled state, everyone at Z, is itself a consistent position: not even "some victim" is forced. Only the demand that the crisis end now requires one. (Computed: section 7.) When guilt is grounded the picture is different: clear the others by evidence and one person comes out guilty by himself — necessary, and not chosen. (Computed: the control in section 7.) What the kernel cannot show is that a crowd can bear an open case. That is a human question, not a logical one. My reading, which the model does not compute: an open case costs everyone a little — the unease of not knowing — instead of costing one person everything.

So there are two honest exits, and neither is a victim: establish where you can, leave it open where you cannot. The price of the open case is to live in the unresolved, without quiet, in anxiety. The price of the scapegoat is a person. The mechanism rests on calling the first price unbearable and the second necessary. Both are lies: the unbearable can be borne, and the necessary victim is only a chosen one.

So if someone sells you an order bought with someone's guilt, ask one question — the cheapest there is, and the most frightening for the seller:

did you check?

Often the answer is no. Unanimous — yes. Certain — yes. Checked — no. And then the peace you are offered is on credit, and one day they will come to collect. For you.

How to check

Run python3 dilemmas/girard.py in the ZTL repository. It prints the nine computed claims above with their three controls, and stops with an error if the kernel answers anything else. The model and its premises are stated at the top of the file: the symmetric crisis is one model of "all against all", and a different model could give a different count. The kernel is described on this site under The rule and The judge.

This essay first appeared in Russian on the author's blog as «Мир в кредит» (July 2026). This version adds the computations, run against the current kernel on 5 November 2026, and was revised after two critical readings: the claims about the model are separated from the reading, the permanence of the debt is stated with its premise, Girard's méconnaissance is made the hinge of the argument, and the way out now begins where the model points — with establishing guilt, not only with abstaining. None of the original claims had to be reversed.

---

Written with AI assistance (Claude, Anthropic): the original essay with Claude Opus 4.8, this English version and the program with Claude Opus 5.5. Idea, editing and responsibility: Vitaly Reznik.