Two rivers: migration and the island-model equilibrium in two differentiated synthetic genetic pools

Classical island-model theory predicts that differentiation between two populations exchanging migrants settles at F_ST of approximately 1 / (4 N m + 1). This paper tests that prediction in the CORGData governed synthetic pool machinery, with two pools of 200 diploid founders sampled from the real gnomAD v4 non-Finnish European and African/African American frequencies of the five-locus MC1R reference panel, propagated under symmetric migration at rates of 0.01, 0.05, and 0.20 (30 replicates). In the frozen tier (20 generations, mean-of-ratios G_ST), all three arms decline below their founder value but fail the calibration criterion, reaching 0.13, 0.17, and 0.22 of the respective predictions. Amendment A-S5 identified a horizon defect and an estimator defect and pre-registered a second tier: fresh replicates run to 100 generations, judged under Nei's ratio-of-averages estimator. The second tier fails in all three arms as well, at 0.20 to 0.28 of prediction; by generation 100 the rare loci are lost from both pools in the majority of replicates, so the long-horizon levels are quasi-stationary values of a decaying system. Migration homogenizes as declared in both tiers; the quantitative calibration of the classical equilibrium is what fails.