Box

A Box question returns one bounding box — the smallest rectangle containing the target — as top, bottom, left, and right edges, each from 0 to 1 measured from the top-left corner, so the same answer is correct at any image size. It takes no criteria: instructions describe what to find and in which image, and may be a phrase like the coffee mug in photo.

Two design details matter. Box needs jev-1.14.0 or jev-preview plus the images header, and it has no not-found value, so an absent target still returns four edges; the documented pattern pairs Box with a Noul presence check and optional per-edge questions when a crop must not cut the target off. TypeSafe marks it dreamer-ready: format, limits, and accuracy can change between releases.