During December 2025, Sovit contributed to the Metta-AI/mettagrid repository by enhancing simulator configuration handling for curriculum training scenarios. Sovit implemented a system in Python that enables updates to configuration invariants during curriculum runs, increasing the flexibility and robustness of AI simulation workflows. The work focused on backend development and configuration management, allowing simulation scenarios to adapt dynamically as training progresses. Sovit also performed benchmarking code cleanup, improving the maintainability and reliability of benchmarking tools. While no major bugs were addressed, the depth of the work lay in delivering a feature that supports more adaptable and maintainable simulation environments.