Edward Landesberg developed and enhanced evaluation calibration tooling for the UKGovernmentBEIS/inspect_evals and inspect_ai repositories over a two-month period. He implemented judge-based evaluation diagnostics to analyze LLM judge reliability, providing policy estimates and confidence intervals that improved the trustworthiness of evaluation reports. Using Python and YAML, Edward expanded the evaluation workflow with a comprehensive tools index, validation guidance, and documentation updates, streamlining judge-based comparisons and calibration processes. His work focused on API development, data analysis, and unit testing, resulting in maintainable, well-documented features that reduced manual validation effort and supported evidence-based decision-making for model-graded scorer calibration.