Implement a library for evaluating attribution bias in LLMs. Developers can use this to stress-test their RAG systems or search-augmented LLMs for demographic or fame-based bias.