The one number VE-5 hinges on: how long does it take to generate a low-res, dense-keyframe proxy per clip? Drop the same clips you use in the editor (esp. a long phone/AI clip + the 11-clip set). Each is transcoded to a 480p proxy via WebCodecs/mediabunny — the same decode→encode pipeline VE-5 would run in the background. Targets: first clip usable <2s, and generation comfortably faster than real-time so it can run in the background while you arrange the timeline.