The Living Gallery begins with owners
Build a small application that can prepare several image studies, replace a selected study, and shut down predictably. Its central problem is not a universal engine architecture. It is keeping scene data, work, results, and presentation alive for exactly the operations that use them.
Draw the owners first: application state owns the current asset revision; an accepted job owns its input payload; a result owns its produced value. Views are temporary access paths, not a second ownership system. Shared ownership is justified when independent operations genuinely need the same lifetime, not simply because a pointer is inconvenient.
Begin with the checksum batch, then build the supplied asynchronous image pipeline. Its workers prepare owned gradient images, and Gallery publishes the current generation while retaining a prior image on failure. The terminal driver saves a real PPM image. The separate Asio experiment teaches coroutine suspension; the PMR experiment renders tile samples. A window, file decoder, cache, and GPU backend remain distinct integrations.
Draw application, asset, job, result, and display boxes. Label every edge owns, borrows, or identifies by value.
