September 2025 (lab-cosmo/pet-mad) delivered core PETMADCalculator enhancements and expanded testing, focused on improving performance, accuracy, and reliability. Key features delivered: 1) Rotational Averaging Refactor and Performance Improvements, enabling batch processing, improved moment accumulation, enhanced Lebedev grid initialization, and refined rotational standard deviation calculations to boost throughput and accuracy. Commits: d9c928237f06ae9f492c5a8562c5c10f86aa3eca; 51784f64ec13ce10d516d51343a17bdb95499e3b. 2) PETMADCalculator Testing Enhancements, expanding coverage for atomic positions, cell rotation, streaming mean/std for scalar energies, forced equivariance checks under rotation, adaptive batch size handling, and unknown-key rejection during accumulations. Commit: cab6000dbeb5236f932f64e5fe795eb8fe281ef6. 3) No explicit bugs fixed were reported this month; nonetheless, the enhanced test suite mitigates risk and improves reliability. Overall impact: improved numerical accuracy and processing speed, heightened reliability in production, and greater developer productivity through clearer abstractions and robust tests. Technologies/skills demonstrated: Python-based numerical optimization, batch processing, Lebedev grid initialization, rotation-invariant computations, and comprehensive test-driven development.

February 2025 monthly summary: Delivered mts_labels_difference API for metatensor, enabling computing set difference between label sets with mapping to original positions. Implemented core in Rust, exposed a C API, and added Julia bindings, with updated docs. This work provides a robust, cross-language capability that improves data analysis workflows and reproducibility.