●builderIf you're integrating LLMs into hiring pipelines, prompt-level bias mitigations are insufficient — name removal at the data layer is the only tested approach that works.
●policyBias generalizes beyond Western English contexts and resists prompt-based fixes, strengthening the case for structural data interventions in regulated hiring workflows.