Law 2 spends four pages telling you not to work with friends. Then the Reversal arrives and concedes there's one thing a friend is better at than anyone you could hire.
Two things, actually, and they're the same thing seen from either end.
A friend will take a risk for you that an employee won't. And when the risk doesn't come off, a friend is the only person the public will believe you didn't sacrifice on purpose.1
That second one is the whole mechanism. It's not that friends make good scapegoats because they're available. It's that friendship is the alibi.
Greene's setup is plain:
A man of power, for example, often has dirty work that has to be done, but for the sake of appearances it is generally preferable to have other people do it for him; friends often do this the best, since their affection for him makes them willing to take chances.1
Take that clause seriously — their affection for him makes them willing to take chances.
An employee running your dirty work does a calculation. What's my exposure, what's my upside, what happens to me if this surfaces? The calculation is correct and it makes them cautious, and caution is exactly the wrong property in this job.
A friend doesn't run the calculation, or runs a corrupted version of it, because the request came wrapped in a relationship. They're not weighing exposure against upside. They're being asked by someone they care about.
So the affection isn't incidental to the usefulness. It's the thing that suppresses the risk assessment, and it's the reason the friend is better at the job than a professional would be.
Now the second half, which is where it gets cold:
if your plans go awry for some reason, you can use a friend as a convenient scapegoat. This "fall of the favorite" was a trick often used by kings and sovereigns: They would let their closest friend at court take the fall for a mistake, since the public would not believe that they would deliberately sacrifice a friend for such a purpose.1
Read the logic in that last clause slowly, because it's genuinely clever and genuinely awful.
The public reasons: nobody would throw a friend under this. Therefore he didn't. Therefore the friend really was to blame.
The closeness of the relationship is what makes the accusation credible. A sacrificed subordinate looks like a sacrificed subordinate — everyone can see the shape of it, and the blame stays where it started. A sacrificed friend looks like a genuine betrayal by the friend, because the alternative reading requires believing something most people won't believe about anyone.
Stay with that for a second, because it's the part that generalises beyond courts.
The technique works because the audience can't imagine doing it. Their inability to picture the act is the load-bearing component — not the king's skill, not the friend's guilt, not the evidence. A crowd that could readily imagine sacrificing a friend would see straight through it.
Which produces an uncomfortable general rule: the more decent your observers, the better this works on them. Deniability is highest where trust is highest.
And it means the technique degrades in exactly the environments you'd expect it to thrive. Somewhere thoroughly cynical, where everyone assumes the worst, the fall of the favourite reads immediately as what it is. Somewhere people still extend good faith, it's close to undetectable.
(Greene tells the "fall of the favourite" twice — here in Law 2's Reversal and again at Law 26, alongside the cat's-paw. The Law 26 telling is folded into The Scapegoat; what's specific to this page is the friend variable — why affection makes both the risk-taking and the alibi work.)
Enrichment, Law 26 (added 2026-08-09). The second telling arrives eleven hundred lines later and gives different reasons for the same move, which is worth setting out because between them they describe a machine neither passage completes.2
Law 2's argument is the one above: the public won't believe you'd sacrifice a friend, so the accusation sticks. Law 26 supplies three more, none of them about friendship:
That third reason quietly inverts the whole transaction. In Law 2 the friend is an asset you spend at a cost. In Law 26 he is a problem you were going to have to solve anyway, and the scapegoating is how you solve two things at once. Greene's Law 2 line — "after you play that card, you have lost your friend forever" — reads as a price. The Law 26 version reads as a disposal you had already scheduled.
And the two accounts of why the accusation lands are not compatible. Law 2: the public can't imagine it, so it must be true. Law 26: the public reasons "why would the king sacrifice his favorite unless he were guilty?" — which is the same inference stated as a calculation rather than a failure of imagination, and it is doing the work in a court where everyone can imagine it perfectly well and is enjoying it. Greene never sets the two side by side, and the audience he needs is a different audience in each telling.
Greene attaches a cost, and the cost is what makes this a real decision rather than a free move:
Of course, after you play that card, you have lost your friend forever.1
Then comes the calibration, which is the most cynical sentence in Law 2:
It is best, then, to reserve the scapegoat role for someone who is close to you but not too close.1
Work out what that instruction requires you to do.
The deniability scales with closeness — the closer the friend, the less believable the sacrifice, the better the alibi. The cost also scales with closeness — the closer the friend, the more you lose when you spend them.
So "close but not too close" is an optimisation problem, and solving it means maintaining a relationship at a deliberately calibrated distance: close enough to be credible as a friend, not close enough that losing them hurts.
That is a person you are cultivating as ammunition. Not a friend you might, in extremity, have to sacrifice — the instruction is prospective. You are being told to select, in advance, someone to be moderately fond of.
And there's a second-order cost Greene doesn't price. The card is one-time-use against that friend, but the play is public. Everyone else at court watched the favourite fall. The next person you ask to take a chance for you has seen how the last one ended, which means the affection that made the mechanism work has been spent across your whole circle, not just with the one person you burned.
Kings could run this repeatedly because courtiers had nowhere else to go. Most people don't have that.
That last point is worth separating out, because it decides who can actually use this.
The whole play depends on a captive audience. Louis XIV's courtiers watched a favourite fall and stayed, because leaving court meant leaving power, land and standing behind. The pool of people willing to take a chance for the king replenished itself, because there was nowhere else to be.
Take away the captivity and the arithmetic collapses. In any setting people can walk out of, the fall of the favourite is a one-time move against your entire circle at once: you burn one friend and inform everyone else simultaneously.
So Greene's "often used by kings and sovereigns" isn't incidental colour. It's the precondition, stated as a historical detail and never flagged as a limit.
Someone above you asks you to do something slightly outside the lines. Not illegal — just the kind of thing that would look bad written down.
They ask you as a friend. That's the part to notice.
Run the check the friendship is suppressing. If this surfaces in six months, whose name is on it? Not who decided — who acted. If the answer is you and the instruction was verbal, you are holding all of the exposure and none of the authority.
Then ask the question that actually settles it: is there a version of this in writing? Not a memo demanding cover — just a normal message confirming what you agreed, sent the way you'd confirm anything else. Just to confirm, you'd like me to go ahead with X.
A request that survives being written down was a real request. A request that gets a phone call in response was a scapegoat assignment, and you've just found out for free.
And from the other side, if you're ever the one asking: notice that wanting a friend for this job is itself the diagnostic. You want them because they won't push back, and they won't push back because they trust you. That's not delegation. The reason they're the right person is the reason you shouldn't ask.
Strongest support. The mechanism is internally coherent and self-explaining — Greene states not just the technique but why the audience falls for it, which is rarer in this book than it should be.
Tension — the Reversal undoes the law. Law 2 argues friends are unreliable because affection can't be trusted. The Reversal argues friends are uniquely useful because affection can be relied on to make them take risks. Both claims are about the same property. Greene needs affection to be unreliable for four pages and reliable for one paragraph, and doesn't notice.
Tension — no evidence is offered. This is the only substantial passage in Law 2 with no named case attached. Michael III, Sung, Talleyrand, Kissinger and Mao all get narratives. The fall of the favourite is asserted as something "often used by kings and sovereigns" with not one king named. Greene will supply cases at Law 26; here the claim floats.
Tension — against the vault's psychology corpus. The instruction to cultivate a relationship at calibrated distance for future expendability sits against essentially everything the vault holds on repair, accountability and secure attachment. Flagged rather than reconciled.
🚩 SINGLE SOURCE · 🚩 SECONDARY WITHOUT PRIMARY — no case, no citation, no named practitioner. [POPULAR SOURCE].
Tension found in the Law 26 enrichment — the promised evidence never arrives. The note above records that Greene "will supply cases at Law 26." He does not. The Law 26 telling is one paragraph, and it names no king, no favourite, and no court — exactly the same evidential vacuum, restated with more mechanism attached.2 Both passages describe a technique "often used by kings and sovereigns" without producing one. The cases Law 26 does supply — Ts'ao Ts'ao's commissariat chief, Cesare's lieutenant, Richelieu's marshal, Roosevelt's secretary — are all office-holders rather than favourites, and none of them turns on the intimacy that both passages claim is the load-bearing part.
Open questions.
Greene's Seven Courtier Types, from The Laws of Human Nature, is written for the person trying to survive proximity to power — read the type, understand what they want, protect yourself.
This page is the same court described by the person at the centre of it, and the two together make an uncomfortable fit.
The later book's implicit promise is that a sufficiently good reader of people can navigate a court safely. This page describes a move that specifically defeats good readers: the scapegoat is chosen because the relationship is genuine, so there's nothing false to detect. The friend isn't misreading the king's affection. The affection is real, and it's real at a level that was selected in advance to be affordable.
What the pairing yields: the courtier's skill of reading sincerity is useless against a threat that doesn't require insincerity. The later Greene's diagnostics are all aimed at detecting a mask, and this move doesn't wear one.
→ psychology · Girard's Scapegoat Mechanism
Girard's account: a community discharges its internal tension onto a victim, and the discharge works because the community genuinely believes in the victim's guilt. The unanimity is not cynical — it's the mechanism's core requirement.
Greene's version keeps the belief and relocates the cynicism to a single operator.
That's the difference worth having. In Girard the whole community is inside the illusion, and nobody is standing outside running it. Here the audience supplies the sincere belief — he wouldn't sacrifice a friend — while one person engineers the conditions that produce it.
So the pairing yields a distinction Girard's frame doesn't draw: the scapegoat mechanism can be operated from outside without ceasing to work from inside. The court's sincerity is not damaged by the king's cynicism; it's the raw material for it. And that explains why the technique is one-time-use, which Girard's version wouldn't predict — the mechanism regenerates fine in Girard, because everyone stays inside. It burns out here because the operator is visible afterward to at least one person, and a mechanism with a witness who knows is no longer a mechanism.
→ business · The Single-Piece Feedback Rule
The business page holds that the useful signal about a person arrives in how they take a single specific correction, and that this is more diagnostic than any amount of stated intention.
Invert it and you get the test this page needs.
Greene tells you the friend is chosen for their willingness to take chances without calculating. The business page's insight is that a person's response to one concrete, low-stakes probe reveals the structure of the relationship faster than any amount of reflection.
Which supplies the workflow above with its mechanism. Just to confirm, you'd like me to go ahead with X is a single-piece probe — small, specific, and it costs nothing to send. The response is the signal, and it's binary: a written confirmation, or a phone call.
The insight neither page reaches alone: you cannot detect a scapegoat assignment by examining the relationship, because the relationship is genuine — you can only detect it by testing whether the request will survive being made legible. Sincerity is not the variable. Documentability is.
Sharpest implication. The fall of the favourite works because the audience is decent, and it costs the operator their friend but spends the credibility of everyone else's affection along with it. The move is priced as one-time-use against one person and is actually one-time-use against a whole court — which is why it belonged to kings, whose courtiers couldn't leave, and why it mostly destroys people who try it anywhere with an exit.
Generative questions.