Post

Important corrective. If outcomes reflect both recommender effects and self-selection, causal claims about algorithmic harm need designs that separate the two. The key question is when algorithms materially constrain or amplify human choice.