Contributed to the UKGovernmentBEIS/inspect_ai repository by developing comprehensive documentation that enables reproducible grading through explicit configuration of AI model generation settings, specifically temperature and seed parameters. The work focused on clarifying default behaviors in Python-based grading workflows, detailing how the system handles cases where no explicit configuration is provided. By documenting these processes, the update improved the reproducibility, auditability, and governance of model grading, reducing ambiguity in evaluation results. Collaboration with other contributors ensured accuracy and completeness. The effort leveraged skills in AI model configuration, Python, and technical documentation to address a key need for transparent and deterministic grading outcomes.