3D Tiles spec question - does "box" define a local CS?

Can a 3D-Tiles boundingVolume define a local coordinate system on its own? Or do you need a transform to go with it?

Asked another way: if the center point of a "box" is non-zero, will the GLTF vertices be interpreted as being relative to this point?

I've been evaluating the FME 3D Tiles converter. The output tileset.json has no "transform" nodes in it, but does have "box" elements that appear to be in ECEF coordinates. And the output .b3dm/gltf data has small local coordinates.

The bounding volume describes, but does not set, where the tiles are.

So if the box is defined to be in a certain position, say on the surface of the globe, then something must be transforming the glTF vertices to end up there. If it’s not a transform in the tileset.json, it must be the RTC_CENTER extension, see: