Numerai, a hedge fund that crowdsources its trading models, publishes its financial training data in a deliberately obfuscated and normalized form. As Numerai’s documentation explains, the features are scrubbed so that participants cannot identify which real-world stocks or variables the columns represent - low values mean bad performance and high values mean good performance, but the underlying names are hidden. This design prevents contributors from trading on the data themselves and makes it harder to overfit to known securities, while still letting a global crowd build and submit machine-learning models. Numerai then combines submissions into a stake-weighted meta model that drives its actual fund trading.