On Endogenous Dominance
How Power Manufactures the Conditions of Its Own Continuation
Dominance does not require an initial hierarchy. A system can contain no globally superior actor and still converge toward extreme asymmetry.
Consider the minimal case. Thirty agents occupy a cyclic interaction system, distributed equally across three types. Each type defeats one and is defeated by another. No transitive ranking exists. No type is globally strongest.
A realized run terminates at 29-1-0.
If the rules contain no strongest actor, where did the dominant actor come from?
The instinct is to search the interaction structure. Look long enough and there is nothing to find: no transitive ranking, no encoded winner. Asking why one type won, as if the answer lies in its properties, is asking the wrong question.
The right question is what the wrong question hides: if the winner is not encoded in the interaction rule, it must be produced by the trajectory through which that rule is realized.
Winner is not generated by rule. Winner is generated by trajectory.
This is the observation this paper tries to account for.
II. Power Is Not Capability
Start with what changes when a simulation runs.
At each moment, the system holds a state: population counts, spatial positions, local densities, the velocity at which the composition is shifting. That state determines which interactions are actually accessible, not just which the nominal rules permit. When an interaction occurs, the state updates. The updated state reshapes which interactions are now accessible. The nominal rules stay fixed throughout. The actual interaction landscape changes with every encounter.
From here, two things need separating that are usually run together.
The first is exposure: whether a given type of interaction actually occurs. This is a function of spatial configuration and population density, where actors are, how many there are, how they move. High exposure does not mean favorable exposure. Rock with twenty potential encounters per hour, eighteen of them with Paper and two with Scissors, has high exposure and poor composition. The quantity of encounters tells you almost nothing; the distribution is what matters.
The second is utility: what the encounter is worth if it occurs. Capability determines what an actor can do in a given encounter. The system state determines what that doing is worth. And utility is not just the immediate payoff from winning or losing. An interaction can have a negative immediate outcome and still be worth pursuing if the state it produces is favorable enough. Fabius Maximus gained little from any single refusal of battle. Its value lay in what repeated refusal did to the strategic environment over time.
Effective power, then, is exposure weighted toward high-utility interactions. An actor is not powerful because it has a strong capability. It is powerful when its capability is likely to be called upon in situations where exercising it matters.
Power is not possession. Power is capability made consequential by state.
III. How Advantage Becomes Recursive
So far this is a description of what happens in each interaction. The interesting question is what happens across interactions.
Suppose an actor exercises its capability and wins. The win updates the system state. That updated state reshapes the effective interaction landscape. If the new landscape makes that actor’s capabilities more valuable than they were before, the actor did not just win. It won in a way that made winning again more likely.
This is recursive advantage: the path modifies the conditions under which the next path is chosen.
The general condition for this to hold is what we might call capability/environment recursive fit: an actor’s capability set is such that exercising it changes the environment in a way that increases the value of that capability set. It does not require that the actor be universally superior. It requires only that the actor’s capabilities and the environmental state be mutually reinforcing in the right direction.
Four mechanisms instantiate this general condition through different operations on the interaction graph.
Exposure accumulation. The actor engineers its encounter geometry toward interactions where its capabilities are valuable. It does not need to improve its capabilities. It needs to be in the right place when the right interactions become available.
Rival-edge contraction. The actor’s interactions progressively foreclose rivals’ access to productive encounters. Rival capability does not disappear from the rules but becomes inaccessible in practice: the nominal edge survives in the nominal graph while vanishing from the effective one.
Sequential state construction. The actor exercises capabilities in a sequence such that each exercise manufactures the environmental conditions under which the next capability becomes valuable, making the portfolio a construction sequence in which order is part of the strategy.
Intermediation capture. The actor becomes the node through which other actors’ capabilities must coordinate. It does not destroy alternatives. It makes routing through itself more efficient than alternatives, until alternatives atrophy from disuse.
These are not four parallel mechanisms of equal type. They are four observed instantiations of the same underlying condition. In any specific case, the question is which one, or which combination, is doing the work.
IV. Two Architectures of Dominance
Standard Oil and J.P. Morgan in 1907 are usually treated as separate historical phenomena. Set next to each other, they illustrate two contrasting architectures through which recursive advantage can become structurally embedded.
Rockefeller: Power by Contraction
The Standard Oil strategy was not primarily to build a better refinery; it was to restructure the graph.
Rockefeller secured rebates from railroads: Standard Oil paid less per barrel to ship than its competitors.¹ This was not a capability improvement in the sense of refining more efficiently. It was a payoff engineering operation. The effective return on every distribution interaction shifted in Standard Oil’s favor and against everyone else’s, without anyone’s capability changing.
From that cost position, Standard Oil could price below what competitors could sustain. Competitors exited and their access to markets contracted. Alternative paths did not merely become less attractive. Some ceased to be economically viable; others disappeared with the competitors capable of exercising them.²
Past victory became structural memory. Reversing this kind of dominance requires reconstructing what was destroyed, which is slow and costly because the capability base that would have used those paths has already dissolved. In the strongest cases, the alternative did not merely go dormant; the capability base required to exercise it ceased to exist.
Morgan: Power by Intermediation
The Panic of 1907 was a coordination failure of a specific kind. The problem was not only how many dollars were needed. The problem was: who is solvent? Who needs rescue? Who will contribute? If I commit capital, will others follow? These questions require a trusted central evaluator, not just liquidity.
J.P. Morgan had spent decades building a capability set positioned for exactly this problem: capital, information, reputation, coordination capacity, access, judgment. In normal conditions, his coordination capability was useful. In systemic panic, credible coordination was worth something else entirely.
Morgan evaluated institutions, organized capital pools, negotiated rescues, pulled bankers and trust-company presidents into collective arrangements.³ Each coordination exercise produced not just less panic but a changed environment in which more actors oriented toward Morgan when coordination needs arose. The more indispensable he became, the more information and leverage flowed to him, which made him more indispensable still.
The distributed network of financial actors began to organize itself around Morgan as a central node. Alternative paths were not destroyed but became increasingly secondary. Morgan made other actors’ capabilities composable through himself.
Rockefeller became powerful by making alternative paths disappear. Morgan became powerful by making alternative paths unnecessary.
The difference matters for durability. Rockefeller’s destruction of alternatives is permanent in a way Morgan’s is not. Paths that are destroyed require reconstruction from scratch. Paths that are merely dormant can in principle be reactivated by some actor choosing to exercise them. But dormancy is not stable. Capabilities needed to use dormant paths decay from non-use. Institutions that stop exercising coordination capacity lose it. What begins as dormant drifts toward atrophied, and from atrophied toward destroyed, if no one keeps the path alive.
The durability of dominance is written not only in the paths an actor controls, but in the condition of the paths everyone else stopped using.
V. Why Winning Is Not Enough
After Rockefeller and Morgan, the framework might seem to imply that recursive advantage compounds indefinitely. Accumulate enough, and the trajectory becomes self-sustaining.
Hannibal runs into a problem with that.
Hannibal Barca crossed the Alps in 218 BC with an army that would go on to win at Trebia, Lake Trasimene, and Cannae. Cannae remains one of the most complete tactical victories in military history: tens of thousands of Roman soldiers killed or captured in a single afternoon.⁴ By any measure of local capability, Hannibal was dominant. He did not win the war.
The capability sequence that could have produced Carthaginian victory required military victory to produce political fragmentation: defeat Rome badly enough in the field, Rome’s Italian allies defect from the coalition, the coalition fragments, Carthage extracts a favorable political settlement. The problem was in the link between the first step and the second.
Allied defections required Rome to be weakened past a threshold where allies saw no viable future in the Roman coalition. Rome’s institutional resilience, Senate structure, manpower reserves, and the political costs allies would incur by defecting all pushed that threshold higher than military defeat alone could reach. Hannibal won the field battles repeatedly, consistently, brilliantly. The state changes those victories produced were real. But they were not the state changes the next step in the sequence needed.
Hannibal could repeatedly dominate an interaction without acquiring equivalent control over the production of subsequent interactions.
This is the distinction between local superiority and trajectory superiority. Local superiority is winning the specific interaction reliably. Trajectory superiority is something different: the capability sequence has high transition reliability, meaning each exercise reliably produces the conditions under which the next capability becomes valuable. Hannibal had extraordinary local superiority and a broken sequence link.
The sequence link, not the capability itself, is the unit of strategic analysis.
Fabius Maximus understood this without having the vocabulary for it. He did not try to match Hannibal in open battle. He did not improve Rome’s tactical capability to the point where it could defeat Hannibal on Hannibal’s terms. Instead he systematically denied Hannibal the encounters in which Hannibal’s capability was valuable: refuse pitched battle, harass supply lines, attrit without confronting. Hannibal’s capability remained intact throughout; what Fabius removed was its opportunity to become causal.
Rome’s sequence (survive, attrit, reconquer Spain, force the final engagement on favorable terms) sacrificed local decisiveness for greater transition reliability across the war. Fabius did not become stronger than Hannibal. He made Hannibal’s strength harder to call upon at the moments where it mattered.
VI. Power Manufactures Its Opposition
There is a feedback loop the framework has described: the actor wins, the environment increasingly favors the actor, the actor wins more. This is recursive advantage.
There is a second feedback loop that runs in the opposite direction, and understanding it explains why recursive advantage does not compound to infinity.
As an actor becomes dominant, its dominance concentrates cost, dependency, and threat somewhere in the system. The actors bearing those costs observe them. The concentration can reprice a class of capabilities that were previously marginal: the capabilities whose function is to constrain the dominant actor. Dominance, in this sense, manufactures its own opposition.
Three mechanisms can arrest recursive dominance, only two of which require an opposing actor.
Self-exhaustion is endogenous decay: it requires no strategic actors at all. The dominant capability consumes the very condition that made it valuable. Rock eliminates all the Scissors. The capability that was dominant becomes less valuable precisely because it succeeded: the prey population is gone. No external actor engineers this; it is a structural consequence of the dominant capability consuming the condition that made it dominant.
Counter-adaptation is endogenous resistance: existing actors observe the dominant trajectory and update their strategies. Standard Oil’s dominance did not reduce the value of competition broadly. It reduced the value of competing individually while dramatically raising the value of collective response. Political coalition became more valuable. Antitrust enforcement became more valuable. The actors did not recover access to the original game. They began investing in capabilities that operate on the rules of the game.
Institutional and rule innovation is endogenous restructuring, and the structurally deepest mechanism. Dominance becomes sufficiently costly that a previously uneconomic countermeasure becomes worth creating. Not an existing actor recovering access to a path they’d lost. Not a dormant edge reactivating. A new actor, capability, or interaction rule enters the system because dominance itself made it worth inventing.
If the original game is Rock, Paper, Scissors and Rock is snowballing, the system does not necessarily ask how Paper regains access to Rock. The system may instead create Dynamite. The new interaction rule becomes viable because dominance itself priced it into existence.
The Morgan case offers a version of this. As more and more systemic coordination ran through Morgan, a capability that had previously been low-value began to look indispensable: public institutional coordination. The dependency itself had become a systemic liability. The creation of the Federal Reserve in 1913 can be examined, not assumed, as a potential institutional response to the risks that reliance on private crisis coordination had made visible.⁵
A trajectory can manufacture both the conditions of its continuation and the capabilities required for its interruption.
VII. The Unit of Strategic Evaluation
The practical implication of this framework is a change in the unit of strategic evaluation.
A capability should not be evaluated only by the value it produces when exercised. It should also be evaluated by the interaction environment it leaves behind. A locally attractive action may consume favorable exposure, strengthen a rival’s counter-capabilities, or produce a state from which the actor has no valuable next move. A weaker action may be strategically superior if it increases the accessibility or value of the actor’s subsequent capabilities.
The relevant question shifts from “what does this action produce?” to “what does this action make easier to do next?”
Before exercising a capability, an actor operating within this framework should be able to identify four things: which interactions the action makes more probable; which capabilities it reprices upward or downward in its own portfolio and the rival’s; which alternative paths it causes rivals to abandon or atrophy; and which counter-capabilities become more valuable precisely because the action succeeds.
That last item is the one most often missed. Hannibal’s intelligence at the tactical level was not matched by equivalent attention to what his victories were making possible for Rome. Each major battle repriced Roman institutional resilience upward, not downward. The state produced by the victory was not one from which the next capability in the sequence became more accessible. The question to ask before any move is not only whether the move works. It is whether the state the move produces is one from which the next move also works.
The same framework changes how a competitive situation is read, not just planned. The prior diagnostic question is who has the superior capabilities. The question this framework poses instead: whose past moves have left the most favorable exposure remaining, whose alternatives have been most contracted, and in which direction is the sequence advantage compounding or eroding? A dominant position and a deteriorating trajectory can coexist, and they look identical from the outside until the sequence breaks.
A strategy is trajectory-superior when its successful execution does four things at once: it preserves access to valuable future interactions, increases the value of capabilities the actor already possesses, raises the cost of rival alternatives, and does not generate counter-capabilities faster than it compounds its own advantage.
VIII. The Anatomy of a Dominant Actor
The simulation that opened this paper ends 29 Rock, 1 Paper, 0 Scissors. The question posed was: if the rules contain no strongest actor, where did the dominant actor come from?
The anatomy of a dominant actor does not feature superior capability in any absolute sense. It features a portfolio where exercising capability produces favorable interactions, which produce state changes, which reprice that capability upward, and that sequence is reliable enough that favorable paths concentrate around the actor while rivals’ alternatives atrophy or are foreclosed.
An actor is dominant to the degree that its past interactions have restructured the effective interaction landscape in its favor, and to the degree that this restructuring is durable against reversion, reactivation, rival reconstruction, or the invention of new rules.
Three pathways raise effective power, and they are not equivalent. Capability improvement makes the actor stronger in a given encounter. Exposure engineering increases the rate at which favorable encounters occur. Payoff engineering changes what each encounter is worth, independent of capability level or encounter frequency. Dominance may combine these pathways, but their sequence matters whenever the exercise of one capability changes the conditions under which another becomes valuable.
What prevents recursive advantage from running to infinity is counter-feedback. Dominance may exhaust the conditions that sustain it, provoke adaptation by existing actors, or make entirely new rules worth creating.
A dominant actor is one whose trajectory has outrun its opposition long enough to reshape the landscape significantly. Not permanently. Not inevitably. Long enough.
No actor begins with guaranteed dominance. Each enters a configuration of interactions, some of which will change the conditions under which subsequent interactions occur. Each successful interaction alters the conditions of the next. Alternatives narrow. Exposure shifts. A trajectory acquires memory.
By the time one actor appeared dominant, the decisive event had already occurred. The trajectory had reconstructed the environment around its capabilities.
Power appeared to belong to the actor. Much of it had been manufactured by the path.
¹ Ida M. Tarbell, The History of the Standard Oil Company (McClure, Phillips, 1904); Ron Chernow, Titan: The Life of John D. Rockefeller, Sr. (Random House, 1998). On the specific rebate-and-drawback mechanism, see Michael Reksulak and William F. Shughart II, “Of Rebates and Drawbacks: The Standard Oil (N.J.) Company and the Railroads,” Review of Industrial Organization 38 (2011): 267–283.
² By the late 1880s Standard Oil controlled approximately 90% of American refineries. Library of Congress, Business History: “Standard Oil Established.” The acquisition pace included 22 of 26 Cleveland competitors within the first three months of 1872 alone.
³ Federal Reserve History, “The Panic of 1907,” federalreservehistory.org. On Morgan’s personal role in determining which institutions received aid, see Carola Frydman, Eric Hilt, and Lily Y. Zhou, “Economic Effects of Runs on Early ‘Shadow Banks,’” Journal of Political Economy 123, no. 4 (2015): 902–940.
⁴ Polybius, Histories III.116–117; Livy, Ab Urbe Condita XXII.49. Ancient sources record Roman losses of 55,000 (Livy) to 70,000 (Polybius); modern estimates align with this range.
⁵ The Federal Reserve Act was signed December 23, 1913. For the causal link to 1907, see Federal Reserve History, “The Panic of 1907.” Multiple political and structural factors shaped the final form of the Act, including debates over public versus private control that persisted through 1913.