All reviewers found the novel conformity score to be an "elegant solution to the problem of heterogeneity in classification setting." Proving valid coverage is difficult in this setting, and the paper leverages recent advances to do exactly that. As R4 states, I agree that the work could do with paper revisions that would open it up to be more accessible to the wider NeurIPS community. At the moment, the work is fairly niche in its own community, e.g., it uses non-ML empirical standards of synthetic to MNIST scale.