There are data in your company that are of commercial value. You can't just give them to anyone. But changed or artificial data sets, as similar to real as possible, can be used for performance testing, algorithms debugging and machine learning. Data in those sets should still contain the necessary amount of their statistical properties, but at the same time they should be anonymized.
The development of ClickHouse requires data sets that give approximation for the data of Yandex.Metrica. Alexey will tell about four different approaches they tried to solve this problem, which one eventually wins and how you can use it.