Four Seconds to Falsify
After four experiments showed the King Wen sequence had no effect on AI training, the research shifted domains. Instead of testing whether the sequence could help a neural network learn, the question became: could it help a simulated state survive?
The Warring States period (475–221 BC) is one of history's richest laboratories for strategic decision-making. Seven kingdoms competed through shifting alliances, betrayals, and military campaigns for over two centuries. Among them, Han — the smallest and weakest — survived for 223 years through diplomatic brilliance rather than military power.
A simplified simulation was built: three states (Qin, Han, Chu) arranged in a triangle, each choosing simultaneously whether to attack, fortify, or form alliances. Han was given the King Wen sequence as a policy prior — using hexagram properties to bias its decisions toward unpredictability.
The experiment ran 100 games per condition across seven different agent configurations. It took under four seconds. Han died in 93–96% of all games, typically within five rounds. The King Wen prior made things worse, not better — Han survived 4–5% of the time with King Wen guidance versus 7–9% with random decisions.
But the failure revealed something more important than any success could have.
The Lips and Teeth
To understand why the triangle killed Han, one must understand how the real Han survived.
Han Fei — a prince of Han and one of China's greatest political philosophers — wrote a memorial to the King of Qin arguing that Han should not be conquered. His argument was not that Han was strong, but that Han was useful.
韓事秦三十餘年，出則為扞蔽，入則為席薦。秦特出銳師取地而韓隨之，怨懸於天下，功歸於強秦。
— 韓非子・存韓
Han has served Qin for over thirty years. Abroad, it has served as a shield and buffer; at home, it has been a mat laid out for Qin's convenience. Whenever Qin dispatches crack troops to seize territory, Han follows behind — the resentment of All-Under-Heaven falls on Han, while the merit accrues to mighty Qin.
Han survived not by being strong but by being indispensable. It absorbed the diplomatic cost of Qin's expansion — other states blamed Han for collaborating, while Qin reaped the territorial gains. This parasitic symbiosis kept Han alive for decades.
This is the 'lips and teeth' doctrine (唇亡齒寒) — if the lips are destroyed, the teeth grow cold. Han served as Qin's buffer against the eastern states. Destroying Han would mean Qin's borders directly touched Zhao, Wei, and Chu — none of whom wanted that exposure.
In a triangle, this logic collapses. Qin and Chu are already adjacent. Han provides no buffer between them. Neither neighbor has any strategic reason to preserve Han — eliminating it simply means one fewer competitor for the survivor.
The Art of Playing Sick
Han's diplomats were masters of strategic ambiguity — signaling weakness and strength simultaneously to preserve options that a genuinely weak state should not have had.
The Zhanguoce records an episode where the King of Han dispatches Zhang Cui to negotiate with Qin. What Zhang Cui does next is a masterclass in diplomatic survival.
張翠稱病，日行一縣。張翠至，甘茂曰：「韓急矣，先生病而來。」張翠曰：「韓未急也，且急矣。」甘茂曰：「秦重國知王也，韓之急緩莫不知。今先生言不急，可乎？」張翠曰：「韓急則折而入於楚矣，臣安敢來？」
— 戰國策・韓策二
Zhang Cui feigned illness and traveled only one county per day. When he arrived, Gan Mao said: 'Han must be desperate — you have come despite your illness.' Zhang Cui said: 'Han is not yet desperate. But it soon will be.' Gan Mao said: 'Qin understands your king perfectly well. Can you really claim Han is not desperate?' Zhang Cui said: 'If Han were truly desperate, it would have broken off and gone over to Chu. How would I dare come here?'
Zhang Cui's slow journey, feigned illness, and calm insistence that Han is 'not yet desperate' all serve one purpose: signaling that Han still has the option of defecting to Chu. This is only credible in a world where Chu is a viable alternative — which requires a multi-state system where defection is possible and meaningful.
Zhang Cui's strategy depends on three conditions that the triangle removes:
First, a credible defection threat. Han can threaten to 'go over to Chu' because Chu is a major power that would benefit from Han's allegiance. In a three-state triangle where Chu is already adjacent to Qin, absorbing Han provides Chu with no new strategic advantage.
Second, information asymmetry. Zhang Cui can feign illness and travel slowly because the diplomatic distance between courts creates space for ambiguity. In a simultaneous-move game with perfect state observation, there is no space for this kind of theater.
Third, time. Zhang Cui's strategy works over weeks of diplomatic travel. The simulation runs 20 rounds with immediate resolution. Han's sophisticated diplomatic techniques require a tempo that the game engine does not provide.
The Strategist's Lament
Su Qin, the great coalition builder, tried to rally Han to the anti-Qin alliance by cataloging its military strengths.
天下之強弓勁弩，皆自韓出。谿子、少府時力、距來，皆射六百步之外。韓卒超足而射，百發不暇止。
— 戰國策・韓策一
All the strongest bows and most powerful crossbows in All-Under-Heaven come from Han. The Xiezi, the Shaofu Shili, and the Julai models all shoot over six hundred paces. Han's soldiers step forward and fire — a hundred shots without pause.
Su Qin was selling a vision of Han as militarily formidable. But his real argument was diplomatic: Han's crossbow technology made it a valuable coalition partner, not a viable independent power. The value existed only within the alliance network.
Zhang Yi, arguing the opposite case, devastated this image with material reality.
韓地險惡，山居，五穀所生，非麥而豆；民之所食，大抵豆飯藿羹；一歲不收，民不厭糟糠；不滿九百里，無二歲之所食。
— 戰國策・韓策一
Han's terrain is rugged and barren, a land of mountains. Its crops are not wheat but beans; its people subsist on bean-rice and bean-leaf broth. One bad harvest and the people cannot fill themselves even on chaff and husks. The territory does not extend nine hundred li and holds no more than two years' food supply.
Both Su Qin and Zhang Yi were right. Han was simultaneously formidable (as a weapons manufacturer and coalition partner) and fragile (as an independent state with no agricultural depth). This duality — strength through relationships, weakness in isolation — is precisely what the three-state model destroys.
In a seven-state world, Su Qin's argument wins: Han's crossbow technology and strategic position make it worth preserving as an ally. In a three-state triangle, Zhang Yi's argument wins: Han is simply too small and too weak to survive between two stronger neighbors with no one else to turn to.
The simulation confirmed Zhang Yi's assessment in 4 seconds flat.
The Minimum Viable Complexity
The three-state experiment was designed as a shortcut — test the King Wen hypothesis on a simple game before investing months in a complex one. The logic seemed sound: if King Wen helps in a simple environment, it will probably help in a complex one. If it fails, the investment was small.
The flaw in this reasoning is that it assumes the hypothesis is independent of the environment's complexity. It is not. The King Wen sequence maps 64 hexagrams to states of change. In a three-state triangle with five actions, the number of meaningfully distinct game situations is in the low hundreds. The hexagrams can barely differentiate between them. In a seven-state game with shifting alliances, intelligence asymmetries, and multi-turn diplomatic arcs, the number of distinct situations is in the tens of thousands — a substrate where 64 hexagrams can index meaningfully.
This parallels a finding from the AI training experiments. When the neural network was too small (4 layers, 3 million parameters), no curriculum ordering made any difference — the model lacked the capacity to benefit from structured learning. Only at larger scales does the ordering of experience begin to matter. The minimum viable complexity for the King Wen hypothesis is not a small model or a small game. It is a system complex enough for 64 distinct patterns to be useful.
The three-state prototype took two hours to build and four seconds to falsify. That is good efficiency — failing fast on the wrong path rather than investing months. But the lesson is not that the King Wen hypothesis is wrong. The lesson is that the hypothesis requires a substrate of sufficient complexity. The Warring States period, with its seven kingdoms, is that substrate. The triangle is not.
What Han Knew
Han Fei, writing from the position of the weakest state, understood something that the simulation makes visceral: survival in a multi-agent environment is not about individual strength. It is about making yourself necessary to others.
The shield-and-mat strategy, the lips-and-teeth doctrine, Zhang Cui's feigned illness, Su Qin's flattery of crossbow technology — all of these are strategies that only work in a rich diplomatic landscape. Remove the landscape, and the strategies become meaningless. Han's genius was never about making the right move. It was about making sure there were enough other players that the right move existed.
The next experiment will build that landscape: seven states, historical adjacencies, asymmetric strengths, and the diplomatic depth that Han's survival required. The King Wen sequence will finally be tested in a world complex enough to need it.
夫韓，小國也，而以應天下四擊，主辱臣苦，上下相與同憂久矣。
— 韓非子・存韓
Han is a small state, yet it has endured attacks from all four directions under Heaven. Its ruler has borne humiliation and its ministers have suffered; ruler and ruled have shared these sorrows together for a long time.
Han Fei's memorial does not argue that Han is strong. It argues that Han's suffering has created a bond between ruler and ruled — a shared experience of endurance that makes Han resilient in ways that raw power cannot measure. This is the quality the King Wen sequence is meant to capture: not strength, but the kind of adaptive resilience that emerges from sustained exposure to adversity.