Persona Conditioning as an Assessor-Sensitivity Probe for LLM-Based IR Evaluation

Large language models (LLMs) are increasingly used as relevance assessors in information retrieval (IR) evaluation, raising questions about how assessor framing affects judgment reliability and downstream system comparison. We study persona conditioning as a diagnostic mechanism for exposing LLM ass...