Possession laws are pretty strict and hard to decode. I wouldn't want to be the test case in court. The idea of "poisoning" a dataset is an interesting theoretical. But in practice, I just want to judge the likelihood that the dataset is poisoned by the presence of images. If it is then there's not much I can do with it.
What amazing work! I am very interested in doing research with Tor and a dataset like this could make my job a heck of a lot easier. I have a legal question though: Are your scrapes text only? Before I work with this dataset, I want to make sure that there's no possibility it contains illegal images (child porn).