Allocative, Representational & QoS Harms

Build a working taxonomy of algorithmic harms. Learn allocative harms (denied loans, jobs, services), representational harms (stereotyping, erasure, demeaning content), quality-of-service harms (worse accuracy for some groups), denigration and dignitary harms, and the link from each harm type to a measurable signal in the model evaluation.