games where players form coalitions and share joint gains — the mathematical foundation for fair cooperation

solution concepts

Shapley value — the unique attribution satisfying efficiency, symmetry, null player, and additivity. each player earns their average marginal contribution across all orderings. in cyber, this distributes focus rewards proportionally to each neuron's causal impact on $\Delta\pi$

core — the set of allocations that no coalition can improve upon. a game has a non-empty core if and only if it is balanced (Bondareva-Shapley theorem). stability: no subgroup has incentive to break away

Nash bargaining — two-player cooperative solution maximizing the product of surplus gains. extends to $n$-player settings via axioms: symmetry, Pareto optimality, independence of irrelevant alternatives, invariance to affine transformations

in cyber

the cybergraph is a continuous cooperative game. neurons form implicit coalitions by contributing cyberlinks in the same epoch. the total value is the free energy reduction $\Delta\mathcal{F}$

probabilistic shapley attribution makes fair attribution tractable at scale — Monte Carlo sampling reduces $O(n!)$ to $O(k \cdot n)$, feasible for $10^6$+ transactions per epoch

implemented as an independent layer: cybernet (inspired by yuma consensus from bittensor). experimentally deployed in space pussy, with cybertensor providing CLI compatibility

see cooperation for evolutionary foundations. see learning incentives for the full reward mechanism

Local Graph