Market design for heterogeneous, compositional data goods

Design markets and mechanisms for datasets whose value is heterogeneous and compositional, including mechanisms that handle interdependent valuations among buyers, standards for provenance and quality certification enabling price discovery without full inspection, and computationally feasible attribution methods for models trained on millions of sources.

Background

Data’s heterogeneity and compositional value create interdependent valuations that standard markets do not handle well. The verification paradox (inspection enables copying) and the intractability of attribution when models train on vast sources impede price discovery and compensation.

Section 6 calls for mechanism design, institutional standards for provenance and certification, and practical attribution approaches, drawing analogies to the historical development of standards and exchanges for grain and oil.