Contributed to the UKGovernmentBEIS/inspect_ai and UKGovernmentBEIS/inspect_evals repositories by developing and enhancing judge-based evaluation calibration tooling and documentation. Focused on Python scripting and API development, the work introduced a diagnostics tool for analyzing LLM judge reliability, providing policy estimates and confidence intervals to improve evaluation accuracy. Enhanced documentation in Markdown and YAML clarified calibration workflows and best practices, supporting maintainability and onboarding. Improvements included code quality updates, type hinting, and dependency management, ensuring robust and reliable evaluation processes. These contributions enabled more trustworthy, calibrated evaluation reports and streamlined judge-based comparisons, reducing manual validation and supporting evidence-based decision-making.