Built on Active Inference and large language models, the platform simulates populations of up to 10,000 agents forming preferences and voting under different rules. Agents are calibrated using real survey data, and their internal representations are independently validated using a 144-dimensional cross-encoder, rather than relying solely on self-report.
Building on links between quadratic funding and quantum formalism identified by Fabinger, Freedman, and Weyl, the project delivers a classical, polynomial-time scoring rule that reproduces key properties of the quantum voting pipeline without requiring quantum hardware. The rule resolves Condorcet cycles, preserves Pareto efficiency and non-dictatorship, and introduces a controlled trade-off in place of strict independence of irrelevant alternatives.
The project tests its methods in the City of Aarhus, Denmark, and QuantumBasel, Switzerland, representing contrasting collective decision scenarios. These pilots compare novel voting rules with existing participatory processes and feed results back into the simulation platform.