{"scheme":"Five spatial groups; three train, one calibration, one test; 150 m control exclusion; no full-control refit","folds":[{"fold":0,"train":177,"calibration":71,"test":52,"minimum_train_test_distance_m":535.3970489272424,"minimum_calibration_test_distance_m":757.1657678474378},{"fold":1,"train":174,"calibration":55,"test":71,"minimum_train_test_distance_m":535.3970489272424,"minimum_calibration_test_distance_m":612.2499489587566},{"fold":2,"train":186,"calibration":59,"test":55,"minimum_train_test_distance_m":535.3970489272424,"minimum_calibration_test_distance_m":612.2499489587566},{"fold":3,"train":178,"calibration":63,"test":59,"minimum_train_test_distance_m":535.3970489272424,"minimum_calibration_test_distance_m":638.6900656813131},{"fold":4,"train":185,"calibration":52,"test":63,"minimum_train_test_distance_m":535.3970489272424,"minimum_calibration_test_distance_m":638.6900656813131}],"controls":300,"heldout_high_count":1,"heldout_high_fraction":0.003333,"heldout_family_consensus_count":1,"old_new_spearman":0.979456,"old_new_high_jaccard":0.771263,"shifted_spearman":0.980218,"shifted_high_jaccard":0.74322,"limitation":"Controls were selected for stable five-year trajectories. This tests relative screening consistency within that selected cohort, not field accuracy or nominal false-positive control. Mean family ranks are not probabilities or p-values. A block shift is a sensitivity check, not independent replication."}