Dataset Watermarks

Leakage in data markets is going to be a challenging issue to fit. We need some way to track leakers of a dataset who attempt to sell it outside the bounds of the data market. Digital watermarks attached to each dataset seem to be a good method for this, but the challenge is that it shouldn't be feasible for new members of the data market to recover un-watermarked versions of the dataset. Need to devise a scheme for this.