Deployed pipeline: SAM + op_sam localizer + within-volume prior + pyramid keep-best — test Dice 0.9200 (this cache: mean 0.7299 / median 0.9063, n=28).
Tiles show deployed-representative slices (the deployed volume-prior + pyramid stage recovers the pre-prior localizer misses; deployed min 0.829).
▮ 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).





