The LLM-r/Changemyview field exp, beyond the ethical issues, has a big stats issue: excluding posts that got deleted, which itself may influenced by treatment. This conditioning on a post-treatment variable violates random assignment & means you can't make causal inferences about treatment effects