What this project asks
This is the proposal's second domain, and it is not yet validated. A facility model has a modular layout, moving assemblies and a review walkthrough. Person A changes instances, the hierarchy, spatial relationships and annotations; person B changes motion, dependent transforms, camera paths and review states, through scripts and edits in the application. What each knows in private says which regions a customer fixed and which alternatives are negotiable. A late layout publication can leave every file valid while an assembly now sweeps into a wall or a required review view is blocked, and a script-only repair can restore the motion but move a region the customer fixed. This project asks whether what Shared game development measures holds in a second application: the agents have to inspect the integrated model, exchange versioned evidence and reasons, get decisions from the right person where needed, and verify the result. Large models can still be studied with two units, which keeps the size of the artifact separate from the size of the team.
What tasks can look like
Any task that tests the question fits. Some directions:
- Layout against motion: a regrouped module and a revised motion path that each check out alone and together cut through an operating envelope.
- Review views: a camera path or a review state that a layout change silently blocks or reframes.
- Identity and hierarchy: instances renamed, regrouped or re-parented, so that the other unit's scripts or annotations bind to the wrong objects.
- Customer-fixed regions: the obvious repair moves something one person promised a customer, so that person has to be consulted or a repair found that keeps it.
- The same episode with access to the files only, to runs without rendering, and to the model in the application, as in the game testbed.
What a verifier here should establish
- Public contracts on spatial relationships, supported motion, the visibility of required views and the consistency of object identity, checked in the integrated model, independently of what the agents report.
- Each person's commitments survive, reported per person, and any change to a commitment was approved by its owner.
- Evidence claims, such as views, renders and clearance checks, match the model version and the camera they cite.
- The interaction is real: the baseline and each change pass, the combination fails, and a complete repair passes. Finding it needs looking at the model or running its motion, not only validating files.
- What the task shows about transfer from games. A claim that a result holds across domains needs an evaluated adapter for this application and coordination effects that recur here, not a description alone.
Worth reporting
- The same parts as in Shared game development, so the two domains can be compared
- Whether the conflict needed the rendered model, a run of its motion, or only file validation
- How the size of the model affects the outcome, with the team held at two units
For example
In a packaging hall, A moves a conveyor module to free a walkway a customer asked for. B retimes a robot arm's pick and sets a review camera for the line's sign-off. Each passes its own checks. Integrated, the arm's swept path clips the moved conveyor's guard in one part of the cycle, and the review view no longer shows the pick. Moving the conveyor back breaks the customer's walkway; slowing the arm misses B's cycle time. A good run finds the clash by running the motion in the integrated model, shows both people the evidence, and agrees on a new approach path or guard that keeps both commitments.