Who is involved
A supplier builds and approves a model release. An operator runs the machines that receive it. When something goes wrong, both need to agree on which release was meant to run and which one did.
What changes in the field
Machines are not always connected, and an update can sit approved for a while before a machine picks it up. That makes the gap between approved and loaded, and the gap in reports, ordinary rather than rare.
What evidence would help
A record per machine of the approved release, the reported loaded release and the time of each report, kept even when a machine goes quiet. That is the same comparison Contour makes in its experiment, applied where reports are less regular.
Where Vraelis is
The autonomous systems application page shows a release-review scenario for machines in the field, illustrated by a synthetic ONNX CPU run. It is not a robot integration and not evidence of physical safety.
