For decades, game theory has explained cooperation as a strategic choice: agents cooperate because the long-term cost of retaliation outweighs the short-term gain of betrayal. Robert Axelrod’s famous 1980 tournament demonstrated that Tit-for-Tat, a strategy that simply mirrors its opponent’s last move, wins against far more sophisticated competitors.
We asked a different question: what if cooperation isn’t a strategic discovery at all, but a geometric property of how stable systems work? To test this, we stripped away everything game-theoretic—payoff matrices, iterated game rules, concepts of “cooperation” and “defection”—and asked whether cooperation would still emerge from nothing but the physics of structural stability.
The Precursor: EXP-002
Before scaling to hundreds of thousands of runs, we needed to show that the engine’s physics could produce game-theoretic outcomes at all. In our initial proof-of-concept (EXP-002), we ran a single deterministic 200-tick simulation pitting eight classical strategies against each other. The agents were pure physics objects, governed entirely by two parameters: force (Lambda) and structural contribution (Gamma). There were no payoff matrices, no iterated game rules, and no concept of “winning.”
EXP-002 utilized an event mutation script (EDMS) of 44 temporal interventions that modeled the unfolding of strategy interactions, such as spiking an agent’s force parameter at a specific tick to simulate retaliation. The physics engine then computed the consequences: the structural isolation of defectors, the exploitation of unconditional cooperators, and the stability dominance of cooperative-retaliatory strategies. All eight established Axelrod dynamics emerged correctly from the engine’s calculations, with Pavlov and both Tit-for-Tat agents occupying the top three positions.
The result was promising, but the EDMS acted as a choreographer. The engine calculated the right outcomes, but the timeline of interactions was explicitly guided. To claim that cooperation is a structural necessity, we needed to remove the training wheels entirely.
The Experiment: EXP-004
EXP-004 eliminated the EDMS completely. Zero pre-scripted events. Instead, we mapped 15 classical strategies to static force (Lambda) and structural contribution (Gamma) parameters, and let the KAIROS physics engine govern all interactions natively.
Each agent operated using an asymmetric utility function that captured three physical assumptions: cooperation costs the giver energy, the receiver benefits from the giver’s structural contribution, and force degrades all interactions. No strategy-specific decision functions, no memory of past interactions, no concept of cooperation or defection—just physics objects navigating a stability landscape.
We swept 1,458 parameter configurations across the full range of the cost-to-benefit ratio, running each configuration 200 times with different random seeds. Total: 291,600 deterministic tournaments.
The Findings
Without game rules, and without a script telling agents when to act, the stability engine reproduced Axelrod’s core results—and revealed structural features of cooperation that game theory treats as assumptions rather than predictions.
Three Regimes, One Parameter
The most commercially relevant finding: a single measurable parameter—the cost-to-benefit ratio of cooperation (α)—perfectly determines which regime a system occupies.
When cooperation is cheap relative to its benefit (α < 0.81), the system lives in abundance: cooperation is individually optimal and no dilemma exists. When cooperation is costly but still collectively beneficial (0.81 < α < ~1.0), the classic Prisoner’s Dilemma emerges from the physics. When cooperation costs more than it returns (α ≥ 1.0), the system enters scarcity: cooperation is dominated and collapse follows.
These regimes are not assumed. They are measured from the simulation output and their boundaries are analytically derivable from the stability equation. For enterprise applications, this means a system’s regime—abundance, dilemma, or scarcity—can be calculated from its structural parameters before running a single simulation.
Pavlov Dominance
In the dilemma regime, the Pavlov strategy (Win-Stay, Lose-Shift) dominated in 99.6% of runs, reproducing the post-Axelrod finding of Nowak and Sigmund (1993) that self-correcting strategies outperform purely reactive ones. Non-exploitative strategies occupied all top-3 positions in over 99% of tournaments.
The intuition is straightforward: Pavlov’s outcome-conditional switching allows it to escape retaliatory spirals that trap simpler reciprocators. Where Tit-for-Tat mirrors the last move forever, Pavlov adjusts based on whether the current approach is working—a structural advantage at scale.
Retaliation Without Escalation
The engine revealed a precise mechanism for effective punishment: the most successful way to respond to a defector is through the withdrawal of structural contribution, not the escalation of force. When a Tit-for-Tat agent retaliates by dropping its contribution to a defector, the defector’s interaction score drops dramatically. But the retaliating agent’s force parameter stays low, preserving all of its other cooperative relationships.
Force escalation, by contrast, is self-defeating. An agent that raises its force (Lambda) to punish a defector degrades all of its interactions—including the cooperative ones that generate its long-run stability. The physics makes this tradeoff precise and calculable.
Why This Matters
The retaliation-without-escalation finding translates directly to domains where the temptation to “hit back harder” is strong. In trade policy, retaliatory tariffs (force escalation) damage the retaliating economy’s other trade relationships. Targeted withdrawal of cooperation—reducing specific trade benefits—punishes the defector without collateral damage. In corporate competition, aggressive pricing wars (force escalation) erode margins for everyone. Selective disengagement from partnerships with bad actors preserves the rest of the network.
The regime structure provides an even more fundamental tool: the ability to diagnose whether a system is in abundance, dilemma, or scarcity from its structural parameters alone. A system in abundance doesn’t need governance mechanisms—cooperation takes care of itself. A system in scarcity can’t be saved by better strategies—the physics are against it. Governance and institutional design matter most in the dilemma regime, where the system could go either way. Knowing which regime you’re in is the first step toward knowing what to do about it.
Conclusion
EXP-004 demonstrates that cooperative dynamics emerge robustly from domain-agnostic stability physics across 291,600 tournaments, 15 strategies, and 1,458 parameter configurations—without event scripting, payoff matrices, or game-theoretic semantics. The geometric interpretation of cooperation proposed in EXP-002, that cooperative dominance is a structural property of stability landscapes rather than a strategic outcome, is strengthened by its persistence across hundreds of thousands of independent tournaments.
These results are a demonstration of compatibility between the stability framework and established game-theoretic findings, not a claim to have replaced game theory. The key structural insight—that cooperation is the valley in a stability landscape, not a choice made by rational agents—is consistent with decades of game-theoretic research but derives from independent physical foundations.
This is the second of three computational studies. The Paperclip Maximizer demonstrates that stability constraints outperform unconstrained optimization in single-agent settings, and The Commons Test extends cooperation dynamics to multi-agent competition, revealing the institutional mechanisms that determine whether shared resources survive or collapse.