Deployed pipeline: SegGPT-retrieve ⊕ op_sam edge-gated keep-best ensemble — test Dice 0.7877 (this cache: mean 0.7843 / median 0.8441, n=30).
▮ green = ground truth ▮ red = deployed prediction. Per-image Dice is burned on each tile; 6 fixed test images, chosen deterministically (spread across the dice-sorted test set).