# Scope and precedents

Read 29 September 2026.

RFS's Computer Week experiments supplied the explicit scene / constructive training-data direction: https://un1crom.github.io/computer-week-2026/ . This implementation does not claim authorship of that direction.

The local library's chapters 66, 68 and 72 explicitly distinguish ordinary editable structure from a research advance; supplied-correct relationships from inferred relationships; and tested intervention families from general state sufficiency. This run uses supplied-correct scene state and a limited action family.

Primary sources checked:

* [Text2Tex, ICCV 2023](https://daveredrum.github.io/Text2Tex/): depth-aware image diffusion, partial texture synthesis, view selection and projection into a shared texture. We read its official project page; we did not reproduce its algorithm. Our flat swatch plus authored UV mapping is a substantially simpler conventional baseline.
* [CasTex, WACV 2026](https://github.com/thecrazymage/CasTex): explicit texture maps with physically based shading. The official repository describes CUDA 12.4 / PyTorch3D dependencies and an Objaverse evaluation workflow. We read the README, not the full paper or implementation. This Mac experiment does not reproduce CasTex and does not compete with its reported results. A generated RGB swatch with hand-set roughness is not inferred PBR material.

No claim that generated textures, UV correspondence, persistent object IDs, typed edit logs or undo are novel. A scene renderer already preserves authored geometry under material changes. The engineering question is whether the complete bounded workflow is operational, inspectable and useful enough to justify a production-scene test. Artist acceptance and lower correction effort are still unmeasured.