Evidence levels in a code-built digital twin

Question

A digital twin is most useful when it is complete, yet most of its parts are not known with equal certainty. How can a model show every part and still be honest about the quality of the evidence behind each one?

Context

The CHS 275 twin is built in Blender from Python. Every part and every fastener is a separate object with a stable part ID. The twin feeds an interactive viewer in which a visitor can select a single part.

Method

Each object is assigned one evidence level when it is built. The level travels with the object through the build, the GLB export and the part registry, so the viewer can show it next to the part.

Evidence

The build uses five levels:

Level	Meaning
measured	Dimensions taken from the car itself
photo	Shape or placement matched against calibrated photographs
catalogue	Taken from parts-catalogue information for this model
published	Taken from published technical material
generic	A plausible stand-in where nothing better is available yet

Findings

• Showing every part and labelling its level is more useful than leaving uncertain parts out. A gap in the model hides the uncertainty instead of showing it.
• A label may only be upgraded when the evidence for the higher level exists. Upgrading a label to make the model look more finished would defeat the purpose.
• Keeping the level on the object itself, rather than in a separate document, means every export and view stays consistent automatically.

Unknowns

• How the levels should combine when an assembly mixes parts of different levels has not been decided.
• No measured comparison has been published yet of how far generic parts deviate from the real ones.

Sources

• CHS 275 project overview (owner documentation)
• CHS 275 twin build documentation (owner documentation)

Related projects

• CHS 275
• Blender AI Production System