Detecting anomalies in a time series is a critical task useful for many systems, from sensor feedback in production lines and water plants to digital feedback from server load and incoming connections, fast and trustworthy results are critical for continued safe operation of these systems. This project aims to develop good performing, trustworthy, and interpretable machine learning methods for time series anomaly detection based on previous work on GNNs and other time series methods using techniques such as conformal prediction. This project also aims to promote reproducibility and trustworthiness through careful examination of the literature. Current work has discovered several major flaws in experimental design and reporting affecting much of the work in time series anomaly detection. These flaws impact the ability to objectively judge the performance of current state of the art methods and should be addressed.