Changes in user behaviors, environmental context, or technical systems can cause data distributions to shift, thereby derailing ML model performance. Yandex recently announced the shifts Challenge to motivate the development of models that are robust to distribution shift and methods to detect shift. As part of this challenge, they have OSS’ed 3 tasks designed to mimic real-world, large-scale industrial applications: weather prediction, machine translation, and vehicle motion prediction. The training and development datasets for these tasks have been released and full evaluation data with ground truth labels will become available on November 1, when the challenge concludes.