Independent Jev reports|Not affiliated with TypeSafe AI
jevmodel.ai
Last updated:

Jev Day 2026 · Independent report

Jev Box: a bounding-box question type on jev-1.14.0

Box returns four edges as fractions of the image, plus confidence. It needs jev-1.14.0, an image, and the images header. A missing target still returns a box.

Compiled by jevmodel.ai. We read the cited posts and documentation. The measurements stay with their authors. We have not rerun them.

Four edges and a confidence value

A Box question asks where something is in an image. The answer is top, bottom, left, and right, each from 0 to 1, measured from the top-left corner. left and right are fractions of width; top and bottom are fractions of height, so the same answer can be scaled to any pixel size. The answer also has a confidence value from 0 to 1.

The question has two fields: type, always "box", and instructions that name the target and the image. Box takes no criteria. Almeida calls it a type in the gray area of decisions, and says it is only on jev-preview. The docs require jev-preview or jev-1.14.0, an image in the request, and TypeSafe-Danger-Zone-Images: true.

Sources:[156] TypeSafe AI[149] Diogo Almeida

Every Box answer is a box

A Box question about something absent still returns four edges. In the same request, the docs ask a Noul whether the target is in the image, and tell callers to ignore the box when that Noul is low. One Box question finds one target. Several different things each get their own Box in the same request. When the image may hold more than one of the same thing, the docs say to describe the one you want, or to add a Choice with none / one / several and split the image if the answer is several.

A request can mix Box with Choice, Score, and Noul. Every question is evaluated against the same state. To crop, multiply each edge by the image width or height in pixels. The docs’ padding example moves each edge out by 0.05 of the image when confidence is below 0.6; both numbers are labelled as values to tune on your own images. There is no recommended threshold yet.

Sources:[156] TypeSafe AI

TypeSafe’s 36-crop edge check

To test whether a box cuts off part of the target, the docs crop to the box and ask one Noul per edge, such as whether the mug continues past the left edge. Edges where the Noul is yes move out, and the question is asked again.

On 36 test crops of paintings, each with a known part of the target cut off, the four edge questions caught 30 of the cuts. A single “Is all of the mug inside the crop?” question caught 26. Drawing the box as an outline on the full image and asking about that caught 20. Those counts are TypeSafe’s, on that 36-crop set.

Sources:[156] TypeSafe AI

Questions

Can jev-1.13.0 answer a Box question?

The Box docs say Box questions need Jev 1.14. Send jev-preview or jev-1.14.0, include an image, and send TypeSafe-Danger-Zone-Images: true.

Sources:[156] TypeSafe AI

What happens if the object is missing from the photo?

Box still returns four edges. The docs pair the Box with a Noul that asks whether the target is in the image, and tell callers to ignore the box when that Noul is low.

Sources:[156] TypeSafe AI

How well did TypeSafe’s crop check catch cut-off targets?

On 36 painting crops with a known cut, four edge questions caught 30, a single all-inside question caught 26, and an outline on the full image caught 20.

Sources:[156] TypeSafe AI

Sources and reading notes

The summaries below are paraphrases. Each link opens the original source, including its surrounding context.

TypeSafe AI

Source checked:

Official

[156] Box: four edges as fractions of the image

A Box question returns top, bottom, left, and right as fractions from 0 to 1, measured from the top-left of the image, plus a confidence value. Box needs an image in the request, the images header, and jev-preview or jev-1.14.0. The type takes no criteria. A Box answer is always a box, so the page pairs it with a Noul that asks whether the target is present. On 36 test crops of paintings with a known cut, four edge questions caught 30 of the cuts, a single all-inside question caught 26, and an outline on the full image caught 20.

Official Box documentation as accessed 15 October 2026. The 36-crop check is TypeSafe’s. There is no recommended confidence threshold; the 0.6 figure in the padding example is labelled as an example to tune.

Diogo Almeida (@CompleteSkeptic)

Published:

Source checked:

X.com

[149] Almeida: Box is a single bounding box on jev-preview

Almeida announces a new question type, box, that returns a single bounding box. He places it in a gray area between decisions, and says it is only on jev-preview.

Third post in the JevDay thread. The Box docs require jev-1.14.0 or jev-preview, an image in the request, and the images header.