The paper proposes a new framework statistical analysis, which pushes randomness to the sampling strategy, but allows the actually data points to be arbitrary. Apart from introducing this model, the paper also studies a mean estimation problem in this framework and provide some approximation guarantees for a proposed estimator. The reviewers (and I) believe this model is very interesting and likely to spur much future work. In addition to the model, the actual results of the paper are also quite strong. An oral presentation would provide the authors with a high-profile forum to share the results with the community.