Writing a transform

An ensemble's pieces have two kinds of body and they are written to differently. Getting this wrong does not error — it silently does nothing, which is the worst failure mode a manipulator can have.

body write to why
an ELEMENT (b3d-destroyable) el.x/y/z, el.rx/ry/rz, el.scale the element OWNS its transform: render() writes the node from those attributes, so a write straight to the node is undone the next time anything re-renders
a NODE (a plain library instance) node.position / rotation / scaling nothing manages it, so the node IS the truth

Scale: the element's, unless it is a stretch

b3d-destroyable's size is the placeholder cube edge length, ignored once library is set (measured: 5.273 wide at scale 1, 2 and 4). tosijs-3d 0.8.1 (#47) added a UNIFORM scale, and 0.10.12 (#105) a per-axis stretch, times scale, which it keeps through renders. scaleElement writes those two. (A per-axis scale written to the node had been reset by the next move or turn, [2, 1, 3] → [1, 1, 1] measured, which is why the attribute exists.)

rx/ry/rz are degrees on an element (yaw/pitch/roll are aliases for ry/rx/rz), matching the format. Babylon nodes are radians, so the node path converts.

The quaternion trap, on our side of the fence

A TransformNode ignores .rotation while it has a rotationQuaternion — and the glTF loader always sets one. That is exactly the bug that made library.instantiate()'s rotation silently inert until tosijs-3d 0.7.0: it wrote .rotation, the quaternion won, and every value produced the model's baked orientation. position worked, which is what made it look wired up.

So rotating a plain node here clears the quaternion first. Skip that and a rotation drag moves nothing, with no error anywhere.