UYVY backend

docs/CURRENT_SYSTEM_ARCHITECTURE.md

@@ -137,11 +137,11 @@ Video input and output are optional edges. `input.backend` and `output.backend`
 
 The input edge writes CPU frames into `InputFrameMailbox`. The current DeckLink backend captures BGRA8 directly where possible, or raw UYVY8 for render-thread GPU decode. The input edge does not call GL, render, preview, screenshot, shader, or output scheduling code.
 
-The output edge consumes completed system-memory frames from `SystemFrameExchange`. The current DeckLink backend schedules those frames to DeckLink. If video output is unavailable, the app continues running render cadence, control, preview, telemetry, and logging.
+The output edge consumes completed system-memory frames from `SystemFrameExchange`. The render thread owns output pixel packing before readback: BGRA8 is read directly, and UYVY8 is packed on the GPU into a half-width RGBA8 target before async PBO readback. DeckLink and NDI output then schedule/send those completed CPU frames without invoking GL or converting pixels. If video output is unavailable, the app continues running render cadence, control, preview, telemetry, and logging.
 
 Runtime state exposes backend-neutral output telemetry through `videoOutput`. Portable fields such as `enabled`, `backend`, and `scheduleFailures` stay at that level; backend-specific counters live under `videoOutput.backendMetrics`.
 
-`PreviewWindowThread` is optional and uses a non-consuming system-memory tap. It paints with Win32/GDI on its own thread and skips preview ticks instead of blocking the frame exchange.
+`PreviewWindowThread` is optional and uses a non-consuming system-memory tap. It paints BGRA8 directly, decodes UYVY8 only for preview presentation, and skips preview ticks instead of blocking the frame exchange.
 
 Screenshot routes are present in the UI/OpenAPI surface but are not implemented in the current native command path yet.
 

docs/FORKING_RENDER_CADENCE_BASE.md

@@ -18,7 +18,7 @@ These parts are the useful base for the fork:
 - `src/video/core`: backend-neutral input/output contracts.
 - `src/video/decklink`, `src/video/ndi`, and `src/video/playout`: concrete video I/O edges and scheduling support.
 - `src/render/thread`: render cadence ownership, readback pumping, runtime render-layer commit point, and metrics.
-- `src/render/readback`: BGRA8 PBO readback and completed-frame publication.
+- `src/render/readback`: BGRA8/UYVY8 PBO readback and completed-frame publication.
 - `src/platform`: hidden GL window/context support.
 - `src/app`: startup, config, video backend factory, runtime layer orchestration, preview, telemetry, and HTTP server hookup.
 - `src/control`, `src/telemetry`, `src/logging`, and `ui`: useful if the new repo still wants a local control surface.

docs/RENDER_CADENCE_GOLDEN_RULES.md

@@ -60,6 +60,8 @@ It must not:
 
 If no completed frame is available, record the miss and keep the ownership boundary intact.
 
+Output pixel-format packing that requires GL belongs in the render-owned readback path before frame publication. Video I/O edges may select supported system-memory formats, but they must not invoke GL or become hidden pixel-conversion renderers.
+
 DeckLink input is also an edge, not a renderer. It may capture/copy the latest supported CPU frame into an input mailbox and update input telemetry, but it must not call GL, schedule output, compile shaders, or drive render cadence. Format decode that requires GL belongs to the render-owned input upload path.
 
 ## 4. Runtime Build Work Produces Artifacts

docs/openapi.yaml

@@ -566,7 +566,7 @@ components:
           properties:
             backend:
               type: string
-              enum: [decklink, none]
+              enum: [decklink, ndi, none]
             device:
               type: string
             resolution:
@@ -578,13 +578,18 @@ components:
           properties:
             backend:
               type: string
-              enum: [decklink, none]
+              enum: [decklink, ndi, none]
             device:
               type: string
             resolution:
               type: string
             frameRate:
               type: string
+            pixelFormat:
+              type: string
+              enum: [auto, bgra8, uyvy8]
+            systemFramePixelFormat:
+              type: string
             keying:
               type: object
               properties:
@@ -752,10 +757,10 @@ components:
           description: Maximum observed render-thread frame draw duration in milliseconds for this process.
         readbackQueueMs:
           type: number
-          description: Most recent duration spent queueing BGRA8 async PBO readback after rendering.
+          description: Most recent duration spent queueing async PBO readback for the selected system-frame pixel format after rendering.
         completedReadbackCopyMs:
           type: number
-          description: Most recent duration spent mapping and copying a completed BGRA8 readback into system-memory frame storage.
+          description: Most recent duration spent mapping and copying a completed readback into system-memory frame storage.
         completedDrops:
           type: number
           description: Number of completed unscheduled system-memory frames dropped so render could reuse the slot.