As far as I can tell, there is no realistic way for a professional game project to author their Scenes entirely via ECS sub-scenes. Third-party assets consistently break using the Graphics for Entities pipeline, and scripts written by tech artists to affect Scene visuals are not easily ported to work as systems/components without help from engineering. Given this is true (happy to be proven wrong here, but have yet to find a serious counter-example), it seems the only reasonable way to author environments for a game that utilizes ECS is to have some kind of baking process that converts relevant Scene collision data into a sub-scene as an authoring post-process step. That way, animations, rendering, and visuals are still controlled by GameObjects and MonoBehaviours, but the simulation aspect of gameplay can still be driven by ECS.
Are there existing tools somewhere in the ECS packages for doing this type of baking? I’ve noticed if I drop all my visual assets into a SubScene their colliders will get automatically converted to ECS physics objects, but this isn’t a functional workflow as the same automatic conversion also breaks the rendering. How can I tell the baking process to find and bake all the colliders in the base scene while leaving my renderer components alone?