Developed a calibration metrics enhancement for the scikit-learn-contrib/MAPIE repository, introducing a classwise flag to the expected_calibration_error function to enable per-class calibration error computation. This addition allows for more granular model evaluation, supporting clearer insights into calibration performance across individual classes. The work involved implementing robust Python code with a focus on data science and statistical analysis, expanding test coverage to validate class label handling and per-class behavior. Documentation and API references were updated to reflect the new functionality, while code quality improvements and consistent testing practices contributed to improved maintainability and reliability of the calibration metrics module.