The Distortion of Prior-Independent b-Matching Mechanisms

A new theoretical paper establishes a fundamental performance limit for ordinal matching mechanisms. Researchers proved no algorithm can achieve a distortion better than approximately 1.582 (e/(e-1)) when allocating items based only on ranked preferences, even with stochastic data. They also designed a mechanism that achieves this optimal bound without needing to know the underlying value distributions. This work shifts analysis from worst-case to expected performance, providing more realistic benchmarks for fairness in AI systems.