One clinic, three policies: benefit and diversity cost of embryo-selection rules in a governed synthetic pool

Four embryo-selection policies are compared inside the CORGData governed synthetic pool under a design frozen in advance: random choice, a noisy phenotypic screen (a labeled proxy), genetic-best choice, and genetic-best choice subject to the allele-frequency cap of the series governance framework. Each family contributes four candidate children per cohort-generation, one carried forward under each policy, over six cohort-generations and 30 replicates. Genetic-best selection buys the largest benefit, 3.9259 founder standard deviations, at the largest allele-frequency displacement, 0.5056. The noisy screen buys approximately half of that benefit at a displacement of 0.2894. No policy, including the random control, sits inside the frozen guardrail, because the displacement bound lies at the drift floor of a 200-family cohort; the frozen greedy cap retains only 0.0287 of the genetic-best benefit. A second tier under Amendment A-S2 replaces the greedy cap with a projected-cohort rule. The projected cap satisfies the cap in 179 of 180 replicate-generations and sits inside the guardrail in all 30 replicates, while retaining 0.0707 of the uncapped benefit. The trait is a synthetic five-locus score; no clinical inference is drawn.