Greene gives an instruction and then, four sentences later, gives its opposite, and does not appear to notice.
First: "it is often wise to choose the most innocent victim possible as a sacrificial goat. Such people will not be powerful enough to fight you."1
Then: "Be careful, however, not to create a martyr… If the scapegoat appears too weak and his punishment too cruel, you may end up the victim of your own device. Sometimes you should find a more powerful scapegoat—one who will elicit less sympathy in the long run."1
So: pick someone weak, because the weak cannot fight back. Do not pick someone too weak, because the weak attract sympathy. The window between those two sentences is where the whole technique lives, and Greene gives no way of finding it.
That gap is not a flaw in the writing. It is the most honest thing in the chapter, because the gap is real and nobody has ever closed it.
The first half of the rule contains a genuinely vicious piece of engineering, and it deserves stating on its own.
"Their naive protests may be seen as protesting too much—may be seen, in other words, as a sign of their guilt."1
Think about what that does to the victim's options. An innocent person, accused, does the thing innocence does: denies loudly, repeatedly, with feeling. And the feeling is the problem — a measured denial reads as composure and a passionate one reads as panic, and panic is what the guilty do.
So the evidence most available to the accused, and the only evidence they can produce at will, is converted into evidence against them by its own intensity. There is no volume at which the denial helps. Quiet reads as resignation, loud reads as guilt.
Greene doesn't invent this. It is the oldest structural cruelty in any accusation system, and it works best against exactly the people the first half of the rule tells you to select: those without standing, without allies, without a prior reputation to be measured against.
Try to state the two conditions precisely and you find they are not about the victim at all.
Weak enough not to fight back is a fact about the victim — their resources, their allies, their position.
Not so weak as to attract sympathy is not a fact about the victim. It is a fact about the audience: what they are willing to watch, how they feel about the accused, whether they have anything invested in the accusation being true.
That is why the rule cannot be reduced to a checklist. The two edges are measured in different units. Greene writes as though there is a single dial marked powerful running from fights back at one end to makes a martyr at the other, and you simply choose a middle setting. There isn't. There are two dials, and only one of them is on the victim.
This is Greene's own escape hatch and it is worth taking seriously, because it points at the better version of the technique.
A powerful scapegoat has resources and allies, which is the cost. What you get in exchange is that the punishment reads as a contest rather than a crushing. Nobody pities a man who could have won.
Which is exactly what Cesare Borgia arranged. De Orco had absolute powers, two years of authority, and a province's worth of enemies. When his body appeared in the piazza the crowd did not think poor man; they thought at last.2 The purple cape was doing part of that work — you do not dress a victim in rank unless you want the audience to see a fall rather than a squashing.
Compare Ts'ao Ts'ao's commissariat chief, who had nothing, protested his innocence to his commander's face, and was told "I know."3 That one worked because nobody outside the tent heard the protest. Had the exchange been public, it is the martyr case exactly.
So the strong-scapegoat variant is not a fallback. It is the safer instrument, and Greene files it as an afterthought.
Read the failure condition once more: "you may end up the victim of your own device. It is important that you remain the victim, the poor leader betrayed by the incompetence of those around you."1
There is only one currency in that sentence, and it is who the room feels sorry for. Not who is guilty, not who is harmed, not what happened. Sympathy is a fixed quantity in Greene's model and the entire operation is a fight over where it settles.
That is an unusually clear statement of what this book thinks a reputation is: not a record, not a set of facts, but a standing allocation of the audience's pity, which can be redirected and must be defended.
It also explains why the technique has no ethical vocabulary anywhere in it. There is nothing to have a vocabulary about. The victim's innocence is not a moral fact in this system; it is a property affecting sympathy flow, which is why the same paragraph can recommend innocence as a selection criterion and warn against it as a risk without any tension being felt.
Fourteen lines above this rule sits a case that breaks it, and Greene prints them on the same page.
Cardinal Richelieu, unable to touch a conspirator protected by the queen mother, tried the man's brother — a marshal of France, with no involvement in the plot — on invented charges and had him executed, to punish the conspirator indirectly and to warn the army.4
Run the checklist. A marshal is not weak, which satisfies the second rule. But he was "sacrific[ed]… innocent" — Greene's own phrase — publicly, on charges everyone could see were manufactured, for something his brother did. There is no reading of do not create a martyr on which that passes.
And the case is presented as a success.
What makes it work, if it worked, is the thing the rule cannot capture: the audience Richelieu cared about had no capacity to act on its sympathy. The officer corps could feel whatever it liked about the marshal; it could not do anything. Sympathy only converts into danger where there is a route from feeling to action.
That is the missing variable, and it is not in the chapter. The martyr risk is not a function of how cruel the punishment looks. It is a function of whether the people who find it cruel can organise.
Something has to be said about a failure, and you have already worked out that it will land better attached to a person than to a process.
You run the calculation without admitting you are running it. Not who is responsible — you know that. The question your mind is actually working on is who can I name without it going wrong for me, and it produces a shortlist in about four seconds.
Watch which names come up. It is never the person with the strongest relationships. It is never the one whose manager will push back, or who has been here long enough to have a constituency. It is the contractor. The person on their second month. The one everybody agrees is talented but who nobody has known long enough to defend.
And here is where the second rule arrives, and you will experience it as decency rather than as strategy. Something in you says not them. It feels like a moral flinch. Check what it is actually responding to — often it is responding to a picture of the room's face when you say the name, and the picture is unflattering to you, not to them.
That flinch is worth keeping even though its origin is self-interested. It is the only functioning instrument you have, and it fires faster than the reasoning does.
Now the harder version. The name that would work perfectly is someone mid-level, competent, moderately liked, with no faction. No sympathy risk, no capacity to retaliate, and — this is the part that makes it feel legitimate — genuinely adjacent to the failure. You would not have to invent a single thing. You would simply have to describe the sequence starting from their step.
Nothing in this chapter will stop you there. Greene's rule is fully satisfied. Which is the point worth taking from the page: the calibration exists to protect the operator, and it is silent, by construction, at exactly the moment when the operator is safe.
Notice when a martyr comes into existence.
Not at the moment of punishment. Afterwards — sometimes long afterwards, when a pattern has accumulated and somebody starts counting, and the individual case gets reread in the light of the ones that followed.
Which makes Greene's rule structurally unusable at the moment of decision. You are being asked to avoid an outcome that is assigned retrospectively by observers you do not control, on the basis of events that have not happened yet. The marshal became a martyr when the shape of Richelieu's method became visible. Nobody could have scored that in 1631.
Greene writes the rule as a pre-flight check. It is closer to a weather forecast made by someone with no instruments.
The strongest support is the protest-reads-as-guilt observation, which is real, well-documented across accusation systems from witch trials to internal investigations, and stated here more crisply than in most places that study it.1
The unresolved tension is that both halves of the rule are correct and they cannot be jointly satisfied by a rule. Weakness prevents retaliation and attracts sympathy. Those are not two settings of one dial; they are two different audiences responding to the same fact. Greene names both and supplies no arbitration, and the cases in his own chapter go both ways.
🚩 SINGLE SOURCE. Every case supporting the calibration is an uncited anecdote from this book. There is no instance in the chapter of the martyr failure actually occurring — the risk is asserted and never demonstrated, which means the rule's second half rests on nothing at all.
Open question. If sympathy only becomes dangerous where the sympathetic can act, then the real selection criterion is does this victim's constituency have a route to organise — a question about the operator's environment rather than the victim's weakness. Does the rule survive being restated that way, and does it then apply to most modern organisations, where such routes exist and are cheap?
Greene's Law 22 Reversal supplies the same warning from the other side, and it is blunter: martyrdom is "selfish and arrogant," and the advice is "leave martyrdom alone." The vault's reading of that passage sits at Martyrdom as an Inexact Tactic, and its finding is that martyrdom is unreliable as a chosen strategy — the effect arrives late, if at all, and the person who chose it is not there to steer it.
Put the two together and something falls out that neither states. Martyrdom is unreliable for the martyr and unpredictable for the operator, for the same reason: it is manufactured by third parties after the fact. Law 22 tells the reader not to bet on it. Law 26 tells the operator to avoid triggering it. Both are describing an outcome nobody in the transaction controls, and neither chapter says so.
History — Mass Terror — The Limitations of Scorched-Earth Counterinsurgency
That page reports an uncomfortable finding: mass terror as counterinsurgency has a documented success rate that is not zero. The useful question turns out to be under what conditions — and the conditions are so narrow and so historically contingent that they supply no usable template.
Greene's martyr rule is the same finding at the scale of a single person, and the counterinsurgency literature has the variable he lacks.
What makes terror work, in the cases where it worked, is the absence of any route from outrage to organisation — no sanctuary, no external patron, no press, no mechanism by which the horrified could combine. Where those existed, brutality reliably generated the resistance it was meant to prevent. Not because it was too cruel in some absolute sense, but because cruelty plus a route equals recruitment.
Transpose that and Greene's rule stops being about the victim's weakness at all. The martyr risk is not produced by the punishment's severity; it is produced by the audience's connectivity. A scapegoat destroyed in front of a scattered, dependent, mutually-unknown group — Ts'ao Ts'ao's camp, Richelieu's officer corps — generates no martyr regardless of cruelty. The same act in front of a group with channels produces one almost regardless of proportion.
Which gives a practical inversion of the chapter's advice, and an unwelcome one for a book about individual technique: the operator's real exposure is a property of the organisation he is standing in, not of the person he selects, and it is largely fixed before he chooses anybody.
History — The Heavy Hand: Rasputin's Curse and the Martyrology
That page describes something specific: almost everyone who attached themselves to Rasputin died badly, and the verdict emerged only when someone started counting backwards through the names. Molchanov sits with his father's death, works back through Iliodor, Hermogen, Damansky, and arrives at what he himself calls a perhaps-superstitious conviction. The pattern was not visible in any single death. It became a verdict through accumulation and narration.
That is exactly how a martyr is made, and it explains why Greene's rule cannot be applied.
Follow the mechanism. Martyrdom is not a property of an event; it is a property of a series, assigned by an observer who lives long enough to count. No one standing in the piazza at Cesena, or in Richelieu's Paris, could have known whether the death in front of them would turn out to be the first entry in a list. The list is written later, by someone with a different question, and the earlier deaths get reread in its light.
So the operator is being asked to avoid producing something that does not exist at the moment of production. He can control the severity, the staging, the victim's rank — every variable the chapter names — and none of them determine the outcome, because the outcome is determined by what he does next, and next, and by whether anybody eventually sets the instances side by side.
Which delivers the one piece of usable advice the chapter never gives, and it is not about calibration at all. The martyr risk is generated by repetition, not by selection. A single harsh act rarely produces a martyr. A method produces one, reliably, the first time an observer notices it is a method.
Sharpest implication. Greene's two rules only look contradictory because both are stated as facts about the victim. Restate them and the contradiction dissolves into two separate questions the operator should be asking about his own environment: can the people who will be appalled reach each other, and has anyone yet noticed that this has happened before? Neither has anything to do with how weak the person you are about to name happens to be — which means the chapter's central selection criterion is measuring the wrong object.
Generative questions