History
History

Hsiang Yu and the Four Spared Escapes

History

Hsiang Yu and the Four Spared Escapes

There is a banquet outside Hsien-yang, sometime after 208 B. C.
developing·concept·1 source··Aug 9, 2026

Hsiang Yu and the Four Spared Escapes

The Sword Dance That Never Started

There is a banquet outside Hsien-yang, sometime after 208 B.C. Hsiang Yu has invited his rival Liu Pang, and his adviser Fan Tseng has arranged the whole thing: in the middle of a celebratory sword dance, Liu's head comes off.

Liu arrives. The banquet proceeds. And Hsiang hesitates over the signal.

By the time he gives it, Liu has read the room and got out.1

Fan Tseng's reaction is one line and it is the best diagnosis in the chapter: "Bah! One cannot plan with a simpleton. Liu Pang will steal your empire yet and make us all his prisoners."2

He was exactly right, and it took ten years.

Four Escapes, Granted

What makes this case unusual is that it is not one failure. It is the same failure four times, by one man, against one opponent, with the outcome correctly predicted after the first.

One. The banquet. The signal comes too late.3

Two. Hsiang marches on Hsien-yang determined to finish it; Liu abandons the city rather than fight at bad odds, and escapes.4

Three. Liu is cornered in a walled city, out of food, army in disarray, and sues for peace. Fan Tseng: "Crush him now! If you let him go again, you will be sorry later." Hsiang decides to be merciful — he wants to bring Liu back alive and "force his former friend to acknowledge him as master." Liu uses the surrender negotiations as a distraction and escapes with a small army.5

Four. Hsiang captures Liu's father and threatens to boil him alive in front of both lines.6

Then the tables turn once, and it ends.

Send Me a Bowl of the Soup

The fourth episode is the one people remember, and it is worth reading precisely.

Hsiang stands the old man up during the fighting and shouts across the troops: "Surrender now, or I shall boil your father alive!"

Liu, calmly: "But we are sworn brothers. So my father is your father also. If you insist on boiling your own father, send me a bowl of the soup!"7

Hsiang backs down.

Look at what just happened. Liu did not call a bluff, and he did not appeal to mercy. He used the fraternal bond between the two of them — the same bond that had been protecting him for four escapes — as an instrument, and turned Hsiang's own sentiment into the reason the threat could not be executed.

Hsiang's respect for the relationship was not merely a weakness Liu survived. It was a lever Liu could pull. That is a different and much sharper claim than the one Greene draws, and it is in his own text.

What Fan Tseng Saw

Before any of it, Fan Tseng had read Liu correctly from a single behavioural change: "This village headman used to be greedy only for riches and women, but since entering the capital he has not been led astray by wealth, wine, or sex. That shows he is aiming high."8

That is a genuine piece of tradecraft and Greene does not stop on it. The diagnostic is not a trait but a delta — the man's appetites did not change because his character improved; they were suspended because they had become expensive relative to something he now wanted more. Sudden discipline in a previously undisciplined person is a statement about their ambition, not their virtue.

Fan Tseng had the correct read, the correct prediction, and no authority. Everything he said came true and none of it mattered.

Analytical Case Study: The Asymmetry That Decided It

Greene's interpretation locates the cause in sentiment: "a fatal sympathy with or respect for the man who, after all, had once been a friend and comrade in arms."9 And he adds the line that makes it a law rather than an anecdote: "But the moment Hsiang made it clear that he intended to do away with Liu, yet failed to accomplish it, he sealed his own doom."10

That sentence is the actual mechanism, and it is about asymmetry rather than about mercy.

After the banquet, both men knew the terms. Hsiang had declared, by his own attempt, that this was a fight to the death — and had then failed to prosecute it. From that moment the relationship was structurally lopsided: Liu was operating under a rule Hsiang had announced and was not himself following.

Which means Liu received something more valuable than four escapes. He received certainty about the game while his opponent was still uncertain about it. Every subsequent decision Liu made was clean; every decision Hsiang made had to be re-argued against a sentiment he had not resolved.

And when the reversal came, Liu's adviser gave the identical counsel Fan Tseng had given — "To let him go would be like rearing a tiger—it will devour you later" — and Liu simply agreed.11 Same advice, same position, opposite response. The war was decided by which of the two men could take his own adviser's counsel.

The Peasant and the Nobleman

Greene's own character sketches set the two men up, and they do more work than he acknowledges.

Hsiang Yu: from the nobility, "large and powerful, given to bouts of violence and temper, a bit dull witted," a mighty warrior who always fought at the head of his troops.12 Liu Pang: peasant stock, never much of a soldier, preferred women and wine to fighting, "something of a scoundrel" — but wily, and with "the ability to recognize the best strategists, keep them as his advisers, and listen to their advice."13

Flag the construction, because this build has now logged it repeatedly. The characterisations are doing the argument's work in advance. A bit dull witted pre-explains every hesitation; listen to their advice pre-explains every success. By the time the banquet arrives the reader already knows how it ends, and the evidence has not been asked to carry anything.

There is also a sorting fact worth stating plainly: Hsiang killed his own commanding officer, Sung Yi, by walking into his tent, proclaiming him a traitor and cutting off his head, then took the army without orders.14 That is a man perfectly capable of decisive killing. His hesitation was specific to Liu, not general — which supports Greene's sentiment reading and destroys the dull witted framing in the same stroke.

The Bounty on His Own Head

The end has a symmetry Greene reports and does not comment on.

Hsiang, alone and on foot, finds a small group of his own retreating soldiers and says: "I hear Liu Pang has offered one thousand pieces of gold and a fief of ten thousand families for my head. Let me do you a favor." Then he cuts his own throat.15

The last decision of his life is a gift to men who could not have taken it from him — the same generosity toward a personal relationship that lost him the war, run one final time.

Liu Pang went on to crush the king of Ch'u, his own former leader, crown himself emperor, and enter history as Han Kao-tsu, founder of the Han dynasty.16

Implementation Workflow

You tried to remove someone and it did not work.

Maybe it was a restructure that got softened, a case you made and then took the edge off, a decision you signalled and then did not execute. Whatever it was, they know now. They saw the attempt.

Sit with the specific thing that just happened, because it is not what it feels like. It does not feel like a loss — nothing visible changed, you are still in position, and you may even feel relief that it did not get ugly.

What actually happened is that you handed them certainty and kept the ambivalence. They now know exactly what the relationship is. You are still deciding. Every move they make from here is clean and every move you make has to be re-litigated against a feeling you have not settled.

That asymmetry is the whole cost, and it is invisible because it does not show up in any position or resource.

So the question is not should I have been more ruthless. It is narrower and more useful: do not declare what you are not going to do. Hsiang's error was not mercy — Liu was merciful too, right up until his adviser spoke. Hsiang's error was making the attempt visible and then not completing it. If you are not certain you will finish something, the move is not to try harder. It is to not signal, to leave the relationship undeclared, and to keep your own ambiguity as an asset rather than handing it to them as information.

And notice the thing Liu did with the soup. If somebody is protecting you out of an old attachment, that attachment is not a shield you are lucky to have. It is a lever, and they will eventually feel you pulling it.

Evidence, Tensions, Open Questions

The strongest evidence is the repetition. Four opportunities, four hesitations, one opponent, an adviser predicting the outcome correctly after the first — that internal consistency is far better support for the sentiment reading than any single episode would be.

Tension: Greene's own text supplies a motive he does not use. Hsiang wanted "to force his former friend to acknowledge him as master."5 That is not mercy or sympathy; it is a demand for recognition, and it explains the hesitations better than pity does. You cannot receive acknowledgement from a corpse. Greene names the motive in a subordinate clause and then attributes the behaviour to sentiment.

Tension: the characterisation pre-decides the argument, and Hsiang's decapitation of Sung Yi refutes the dull witted frame from inside the same chapter.

Tension: the case is used to prove a law it only half fits. Law 15 says crush your enemy totally. Hsiang's problem was arguably never insufficient ruthlessness — it was inconsistency: maximal violence toward Sung Yi and the city of Hsien-yang, repeated restraint toward one man. A law about degree cannot address a failure of consistency.

🚩 SINGLE SOURCE · 🚩 SECONDARY WITHOUT PRIMARY — no source is named. The dialogue, including the soup line and Fan Tseng's outburst, is given verbatim; the material derives ultimately from Sima Qian's Shiji, which Greene neither cites nor mentions. [POPULAR SOURCE]

Open questions. Was Liu's soup answer improvised or prepared — and does it matter, given that it required an accurate model of what Hsiang could not bring himself to do? And the counterfactual the chapter forecloses: if Hsiang had killed Liu at the banquet, does the fraternal bond that produced four escapes also predict that his army holds together afterwards, or was the sentiment load-bearing for his own coalition?

Author Tensions & Convergences

The vault already holds this war from a scholarly source, and the divergence is instructive. Stalemate Resolution — Liu Pang and Hsiang Yu comes from Sawyer's Chinese primary-source corpus and covers a different episode of the same conflict. Two accounts of one war, one popular and undated, one scholarly and sourced — and Greene's does not cite the tradition his own material comes from.

That is worth naming as a general condition of this build rather than a complaint about one page: where the vault holds both, Greene's version is consistently the more vivid and the less anchored, and the two should be read together rather than either being treated as the record.

Within Law 15 itself, this case is the chapter's best and it undercuts the chapter's closing qualification. Greene ends by saying the law is "not, of course, a question of murder, it is a question of banishment."17 Hsiang Yu's failure was precisely a failure to murder, four times, and no available banishment would have helped — Liu had an army and a countryside. The flagship case requires the reading the qualification disowns.

Cross-Domain Handshakes

Behavioral mechanics — Necessary vs Magnanimous Compassion

Siu's distinction: a banker writing a charity cheque is performing magnanimous compassion — it extends beyond what the role requires. A president signing a welfare bill is performing necessary compassion — the role's continued operation requires it. "The master of power always knows the optimal mix for any given move."

Run Hsiang Yu through that and he does not fit either category, which is the useful result.

His mercy was not magnanimous: it bought no reputation, was witnessed by almost nobody, and he did not want it seen. It was not necessary: nothing in his position as a general required sparing a rival, and his own adviser said so four times. The typology has no slot for compassion that is neither role-required nor reputationally productive — mercy that is simply an involuntary property of the person extending it.

And that gap matters, because Siu's frame is built for an operator choosing a mix, which presumes the choice is available. Hsiang's case says the interesting failures are the ones where it is not. He was not calculating badly; he was structurally unable to execute against this one man while being fully able to behead his own commanding officer in a tent. A doctrine of optimal mixes cannot help someone whose constraint is that a specific relationship is not on the menu — and those constraints are invisible until the moment they bind, including to the person who has them.

Psychology — Status Hierarchies

That page's central puzzle is that status itself is rewarding, independent of what it accesses. Dominance produces no direct benefit; the benefit comes through food, mates, allies. Yet the feeling of status persists as valuable even when the material benefits are absent — high status with modest income beats identical income with low status, measurably.

Now put Hsiang's stated motive next to it: he wanted to bring Liu back alive and force his former friend to acknowledge him as master.

That is the puzzle in operational form, and it prices it. Hsiang was already winning materially. What he did not have was the thing the status page says is separately rewarding — the acknowledgement — and acknowledgement has one property that resources do not: it can only be supplied by a living rival, and only by that specific rival. Killing Liu would have delivered every material benefit and destroyed the only source of the thing he actually wanted.

So this is not sentiment failing to override strategy. It is two goals that the chapter treats as one, pulling in opposite directions, and Greene has no category for the second. Which generalises past the case: any conflict where the prize includes being recognised as the winner is structurally unable to obey Law 15, because total victory removes the witness whose recognition is the point. That is the condition of most rivalries between peers, and none of the chapter's cases where crushing succeeded — Wu, Camillus, Chiang's opponents — involved a victor who wanted anything from the defeated party afterwards.

The Live Edge

Sharpest implication. The damage was not done by the mercy. It was done by declaring an intention and then not executing it, which transfers certainty to your opponent while you keep the ambivalence. Ambiguity is an asset right up to the moment you announce which way you are leaning — and Hsiang announced it with a sword dance he then failed to signal.

Generative questions.

  • Liu used the fraternal bond as a lever, not a shield. Is protective sentiment in a rival always available to be pulled that way, and does the person relying on it ever get to know before it is used?
  • Fan Tseng was right four times and had no authority. What is the failure here — the adviser's, the ruler's, or a structural property of advisory positions where being correct and being heeded are unconnected?
  • If the prize includes acknowledgement, Law 15 is unfollowable, because victory destroys the witness. How much of the book's advice quietly assumes the operator wants nothing from the loser afterwards?

Connected Concepts

Footnotes

domainHistory
developing
sources1
complexity
createdAug 9, 2026
inbound links8
next in Robert Greene
Empress Wu Chao
In 654 the empress of China came to visit Wu Chao and her newborn child.