An enterprise is deploying a Claude-based system for employee performance evaluations. Legal counsel flags concerns about algorithmic transparency requirements under emerging AI regulations. An architect must design transparency artifacts to support regulatory compliance. Which documentation approach most directly satisfies AI transparency requirements for this use case?

D is correct because AI transparency regulations (EU AI Act, emerging US state laws) require documentation of: what the system does, what data it uses, known limitations and biases, performance across demographic groups (critical for employment decisions), and human review processes. A model card that covers all these dimensions satisfies the substance of transparency requirements. A is wrong because publishing the system prompt reveals implementation details that may be proprietary and does not address the substantive transparency requirements — regulators and affected employees want to know about bias testing and appeal processes, not implementation prompts. C is wrong because a general description of how AI works is too abstract — regulators require specifics about this particular system's evaluation methodology, known failure modes, and demographic performance differences. B is wrong because API audit logs document the technical execution of individual evaluations but do not fulfill the systemic transparency requirements about overall system design, bias testing, and oversight procedures.