Worked on the elastic/elasticsearch repository to enhance the robustness of machine learning anomaly detection, focusing on spike detection in zero-variance datasets. Addressed a bug in the ChangePointDetector by refining the KDE evaluate method so it accurately returns ValueAndMagnitude objects when bandwidth is zero, which previously led to missed spikes in homogeneous data. Added comprehensive unit and integration tests to validate the fix and prevent future regressions. This work, implemented in Java and leveraging machine learning and testing expertise, improved the reliability of anomaly detection in production environments by reducing both missed alerts and false positives in critical monitoring paths.