That's why we're working on a ranking/feedback system (a la piratebay) so that users can see other user comments on the datasets. Regarding decentralization - we are starting with centralized search, and download times are much better for popular datasets by downloading in parallel from many peers. We are also using a congestion control protocol (Ledbat) that is lower priority than TCP - you will not notice it using your b/w to share data if you are downloading/uploading using TCP.