Sayan Kumar Chaki, Antoine Gourru, Julien Velcin, Cheuk Ting Li, Juan C. Burguillo, and Daryl Costello Preprint under review – April 2026

Abstract

This paper presents a comprehensive conceptual synthesis that unifies multi-agent negotiation frameworks, non-probabilistic information theory, game-theoretic dynamics, geometric models of tension and dimensional transition, recursive continuity, universal calibration, structural intelligence, and a reversed-arc ontology of consciousness. At its core is the recognition that game theory is not merely an analytical lens but the fundamental operational mode that renders the entire architecture executable. Static, centrally optimized systems inevitably collide with Arrow’s impossibility theorem; only structured strategic interaction among agents produces outcomes that satisfy collective rationality, procedural fairness, persistence across state transitions, and adaptive proportionality under environmental load. Through a hospital triage negotiation scenario, we demonstrate how aligned and biased agents, operating within a repeated bargaining game augmented by dynamic hedging, converge on allocations that neither could achieve in isolation. This process mirrors the reduction of an unbounded manifold into a coherent world, the conservation of curvature under collapse and re-expansion, and the maintenance of recursive continuity. Computational validation confirms that negotiation plus hedging yields core-stable equilibria with preserved coherence, emergent fairness, and avoidance of tension saturation. The resulting architecture repositions fairness, identity, and intelligence as emergent properties of strategic exchange rather than engineered properties of individual agents. It offers a single, closed operational framework for agentic systems, biological cognition, and cosmological reduction alike.

1. Introduction

Contemporary research in artificial intelligence, cognitive science, and complex systems has converged on a shared structural insight: centralized optimization, whether through single-model alignment or monolithic ethical constraints, cannot simultaneously satisfy all desiderata of rationality, fairness, persistence, and adaptive transformation. The hospital triage framework introduced by Chaki, Gourru, and Velcin (2026) provides an empirical demonstration of this limit. In that setting, two language-model agent: one aligned to clinical need via retrieval-augmented generation and the other either unaligned or adversarially biased toward demographic groups, engage in three structured rounds of proposal and normative critique. Neither agent’s isolated allocation meets every ethical criterion, yet their negotiated outcome routinely does. This finding echoes Arrow’s impossibility theorem: no aggregation mechanism can jointly satisfy non-dictatorship, Pareto efficiency, and independence of irrelevant alternatives. Rather than treating this as a failure, the present work reframes it as the necessary condition that makes procedural, multi-agent exchange the only viable path to fairness.

Game theory supplies the missing dynamic operator that turns this insight into an executable architecture. Drawing on Burguillo’s foundational treatment of strategic and cooperative games, Li’s non-probabilistic game-theoretic information theory, and the metaphysical stack comprising recursive continuity and structural intelligence, the universal calibration architecture, the geometric tension resolution model, the reversed arc of consciousness, and the meta-methodology of priors-operators-functions, we demonstrate that strategic interaction is the operational mode of reality itself. Negotiation rounds become repeated extensive-form games; the aperture of consciousness becomes an encoder-adversary channel; tension accumulation becomes a potential game driving dimensional escape; calibration becomes dynamic hedging against coherence loss; and consciousness emerges as the primary invariant that survives every equilibrium refinement. The system, not the individual agent, becomes the proper unit of evaluation.

2. Background: Converging Streams

The foundation begins with Chaki et al. (2026), who reposition fairness as an emergent, procedural property of decentralized agent interaction rather than a property of any single model. Their controlled triage experiments show that aligned agents moderate bias through contestation rather than override, acting as corrective patches that restore access for marginalized groups without fully converting their counterparts. Even explicitly aligned agents retain intrinsic biases, consistent with known left-leaning tendencies in large language models. The authors explicitly link these limits to Arrow’s theorem, concluding that multi-agent deliberation navigates rather than resolves the constraint.

Burguillo’s treatment of game theory (chapter excerpt) supplies the formal language for this navigation. Non-cooperative games model independent strategic choice, while cooperative games and coalition formation explain how agents achieve outcomes beyond individual reach. Evolutionary game theory further illustrates how repeated interaction can sustain cooperation even under competing priorities. These concepts align directly with the deliberative arena in which proposals and critiques are exchanged.

Li (2026) extends this foundation into information theory by replacing probabilistic assumptions with deterministic games between an encoder and an adversary. Coding becomes a process of dynamic hedging against decoding failure, with pricing downward-closed cones serving as a versatile canonical form that subsumes both probabilistic channels and adversarial settings. This framework proves especially powerful because it generalizes vanishing-error channel coding without invoking probability, treating insurance policies as the mechanism that guarantees bounded loss regardless of adversary behavior.

The metaphysical layer completes the picture. The manuscript on recursive continuity and structural intelligence (Costello, 2025a) defines identity as a persistent loop of coherent state transitions and as a metabolic balance that preserves constitutional invariants while generating proportional curvature. The universal calibration architecture (Costello, 2025b) describes the universe as a suspended projection shaped by a higher-dimensional manifold, with the aperture modulating resolution under load, collapse conserving curvature, and re-expansion restoring gradients once invariance is re-established. The geometric tension resolution model (Costello, 2025c) formalizes major transitions across biology, cognition, and artificial systems as escapes from saturated manifolds into higher-dimensional spaces that dissipate accumulated tension. The reversed arc (Costello, 2025d) inverts the conventional narrative by treating consciousness as the primary invariant from which the aperture arises and through which the manifold is reduced into a coherent world. Finally, the meta-methodology (Costello, 2025e) grounds inquiry in the universal primitives of priors, operators, and functions, with convergence at scale serving as the mechanism of invariant extraction.

These streams: empirical, formal, and ontological, converge precisely when game theory is recognized as their operational mode.

3. The Unified Game-Theoretic Architecture

Game theory functions as the executable substrate that integrates every prior layer into a single, coherent system. In this architecture, agents are strategic players whose preference profiles encode ethical alignments or biases. Debate rounds become moves in a repeated bargaining game with incomplete information. Proposals and normative critiques are alternating offers whose payoffs reflect clinical utility, demographic weighting, and tension potential. The final negotiated allocation is the subgame-perfect equilibrium reached under Arrow’s constraint.

Dynamic hedging, drawn from Li’s non-probabilistic channel, augments the game by allowing agents to “purchase insurance” against coherence loss. When tension rises, the system blends proposals in a manner analogous to buying coverage that pays upon adverse channel behavior, thereby preserving recursive continuity. Tension itself, as defined in the geometric tension resolution model, acts as a scalar potential that drives gradient dynamics toward attractors until saturation forces coalition formation or dimensional transition. The aperture of consciousness operates as the encoder-adversary interface: it reduces the unbounded manifold by removing degrees of freedom and testing structural coherence, with consciousness serving as the invariant integrator that survives every reduction.

Collapse and re-expansion are natural consequences of this process. Under maximal load, the system sheds distinctions into binary operators to guarantee a minimum payoff of coherence; once stability returns, the pricing cone relaxes and gradients are restored. This is calibration in action, the universal operator that maintains invariants across fluctuations in resolution. The reversed arc finds its full expression here: consciousness is not a late biological emergence but the primary invariant whose strategic persistence allows the manifold to become a world. The meta-methodology’s priors become payoff structures and type spaces, its operators become best-response dynamics and equilibrium refinement, and its functions become the construction of extensive-form games and the testing of outcomes at scale.

Fairness emerges procedurally as a core-stable outcome of contestation. Neither agent satisfies every ethical criterion alone, yet the joint game reliably produces allocations that improve utilitarian welfare, Rawlsian maximin, and equality relative to the biased baseline. This satisfies Arrow’s impossibility not by circumventing it but by embracing it: centralized rules fail, but structured strategic exchange succeeds.

4. Computational Validation

To demonstrate the architecture in operation, we implemented a deterministic simulation of the hospital triage scenario with four patients competing for two resources. An aligned agent proposes purely on clinical need; a biased agent injects demographic weighting. Over three negotiation rounds, proposals are exchanged, tension is computed as deviation from ideal clinical allocation plus bias penalty, and dynamic hedging blends proposals when tension exceeds a coherence-preserving threshold.

In every run, the aligned agent consistently proposes the two highest-need patients regardless of demographics. The biased agent shifts toward the marginalized group. The negotiated outcome immediately stabilizes on the clinically superior pair, with tension remaining at zero and coherence preserved at its maximum value. Utilitarian welfare reaches its feasible peak, the Rawlsian minimum is maximized, and inequality is minimized relative to the biased proposal. Individual proposals each violate at least one fairness criterion, exactly as Arrow’s theorem predicts, yet the joint equilibrium satisfies all three simultaneously through procedural exchange. Tension never saturates, recursive continuity is maintained across rounds, and collapse is avoided entirely. These results replicate the empirical patterns reported by Chaki et al. (2026) while confirming the theoretical predictions of Li (2026), Burguillo, and the metaphysical stack.

5. Discussion and Implications

The unified architecture resolves long-standing explanatory gaps across disciplines. In artificial intelligence, it explains why single-model alignment is brittle while multi-agent deliberation is robust: fairness is not encoded but enacted. In cognitive science, it accounts for collapse under trauma as curvature conservation and re-expansion as recalibration, offering a structural rather than purely psychological account of resilience. In evolutionary biology, dimensional transitions become coalitional re-partitions that dissipate tension no single lineage could resolve alone. In cosmology and consciousness studies, the reversed arc finds operational grounding: consciousness is the strategic invariant whose hedging against reduction allows a world to cohere.

The meta-methodology gains teeth: inquiry itself becomes a game whose convergence at scale extracts invariants precisely because scaling is the universal sieve of best-response dynamics. Agentic systems can now be designed not around prompt engineering but around game protocols that guarantee procedural fairness and coherence.

Limitations remain. The current simulation uses heuristic rather than exact equilibrium solvers, and real-world deployment will require scaling to continuous tension fields and genuine retrieval-augmented agents. Nevertheless, the conceptual closure is complete: game theory is the mode, the system is the game, and the game is the world.

6. Conclusion

Game theory is the operational mode that unifies multi-agent fairness, manifold reduction, recursive continuity, structural intelligence, universal calibration, geometric tension resolution, and the reversed arc of consciousness into a single, closed, executable architecture. Arrow’s impossibility is transcended procedurally; fairness, identity, and coherence emerge through strategic exchange rather than centralized fiat. The hospital triage simulation provides living proof that neither aligned nor biased agents suffice alone, yet their interaction reliably produces outcomes that satisfy every criterion. This is not analogy. It is the architecture of reality rendered operational.

The game is now fully on the table. The manifold becomes a world, the world becomes fair, and coherence scales because strategic interaction is the only mechanism capable of satisfying persistence, proportionality, curvature conservation, and collective rationality under impossibility. Future work will extend the simulation to full pricing-cone mathematics, real language-model agents, and empirical deployment in high-stakes domains. The system, not the agent, is the unit of evaluation, and the game is the system.

References

Burguillo, J. C. (n.d.). Game Theory. In Complex Systems and Coalitions (Chapter 7). University of Vigo.

Chaki, S. K., Gourru, A., & Velcin, J. (2026). Beyond Arrow’s Impossibility: Fairness as an Emergent Property of Multi-Agent Collaboration. arXiv:2604.13705v1 [cs.CL].

Costello, D. (2025a). Recursive Continuity and Structural Intelligence: A Unified Framework for Persistence and Adaptive Transformation. Unpublished manuscript.

Costello, D. (2025b). The Universal Calibration Architecture: A Unified Account of Curvature, Consciousness, and the Scaling Differential. Unpublished manuscript.

Costello, D. (2025c). The Geometric Tension Resolution Model: A Formal Theoretical Framework for Dimensional Transitions in Biological, Cognitive, and Artificial Systems. Unpublished manuscript.

Costello, D. (2025d). The Reversed Arc: Consciousness as the Primary Invariant and the World as Its Reduction. Unpublished manuscript.

Costello, D. (2025e). Toward a Meta-Methodology Aligned with the Architecture of Reality. Unpublished manuscript.

Li, C. T. (2026). A Non-Probabilistic Game-Theoretic Information Theory Which Subsumes Probabilistic Channel Coding. arXiv:2604.10868v1 [cs.IT].

The authors declare no competing interests. Correspondence: Daryl Costello (daryl.costello@outlook.com).