The problem of data leakage in machine learning models is a critical issue that can have disastrous effects on model quality if not handled correctly. When training models to predict future events, it's essential to avoid using information from the future in the training dataset to prevent paradoxes and inaccurate predictions. To solve this problem, various approaches such as the "log and wait" method, backfilling historical feature values, snapshot-based time travel, and continuous time travel are discussed. These methods can help mitigate data leakage issues, but they also come with their own set of challenges and limitations. Tecton is a data platform that enables easy construction of high-quality training datasets by performing point-in-time joins to deliver the right values at the right time, making it simpler for modelers to focus on modeling while ensuring accurate training data sets.