CoreVideo Pro

CoreVideo Pro is a native desktop production studio for Windows and macOS for building polished live shows, recordings, and streams from Zoom participants, cameras, screen capture and media: Magic Scene auto-layout, lower-thirds, graphics, audio leveling, and multi-destination output, in a single operator console.

Work order lives in one place: docs/BACKLOG.md. This README is as-built status, not a queue.

WinUI 3 shell (.NET 9, native XAML)  native-shell/CoreVideoPro.WinUI  ← Windows UI
SwiftUI shell                      mac-shell/                       ← macOS UI
  └─ typed command/snapshot          JSON-line over child stdin/stdout; Windows DXGI
       protocol (JSON-line)          shared textures for program/preview/multiview
       └─ C++ media core             native/   (real-time pixels/PCM/transport)
          ├─ Zoom engine subprocess  native/zoom-engine/  (Zoom SDK → I420 shared memory)
          ├─ D3D11 / Metal compositor · audio mixer/DSP
          ├─ Media Foundation / AVFoundation recorder · output adapters
          └─ virtual-camera publisher · diagnostics / support bundles

Virtual camera (out-of-process)       native/virtualcam-dll/  → corevideo-virtualcam.dll
  loaded by the Windows Frame Server; reads the program from cross-session shared memory
  and presents "CoreVideo Pro Camera" to Zoom / Teams / OBS at 1080p60

Shared generated lifecycle models    contracts/   (schema + cross-language wire fixtures)

The native core owns real-time media. Shells send production intent over child-process stdin/stdout and consume snapshots; GPU surfaces and shared-memory media use separate transports. Windows uses D3D11 and Media Foundation; macOS uses Metal and AVFoundation.

The additive lifecycle schema generates C++, C#, and Swift models and runtime validators. Golden wire fixtures run across language suites. Legacy scene/audio/capture and production command/capability protocols still have handwritten mirrors; see the coverage inventory and ownership map. The native hello profile now reports capabilities from constructed adapters. capabilityStates distinguishes available, omitted, and failed-to-construct paths; the legacy capabilities list contains only available paths. An available path still needs live endpoint acceptance (#617). Unknown commands in a production-sync batch can be dropped (#616).

Recording intent, verified media activity, and file finalization are distinct. A Stop acknowledgement does not certify a playable finalized file. Zoom join/auth runs on a cancellable worker while the bounded command mailbox continues serving operator requests. See command semantics.

The default in-container build is stub-first (-DCOREVIDEO_STUB=ON, -DCOREVIDEO_ENABLE_DEV_ADAPTERS=OFF): every capability has a deterministic synthetic implementation so contracts and tests run with no hardware. Real Zoom SDK, GPU, encoder, and hardware-transport code lives behind COREVIDEO_ENABLE_DEV_ADAPTERS plus a per-feature COREVIDEO_WITH_* flag. Windows and macOS builds select their platform adapters; vendor integrations require the corresponding SDKs and runtimes.

Windows merge CI runs the stub core plus the WinUI shell, and a separate native-production-windows job compiles the D3D11 and Media Foundation adapters. The real Zoom SDK is still not compiled in CI (#618): Zoom, WASAPI, UVC, WGC and the virtual camera are proven only on local and packaged builds. macOS CI builds the real Metal/AVFoundation configuration.

As built on main, October 23, 2026. Three states, and only three:

• Proven — shipped in a beta and measured on a real run, with the evidence named.
• Shipped, not live-verified — in the beta, covered by unit or hosted tests, never measured on a real show or installed run.
• Not shipped — code exists behind a flag or on a branch, or only a probe exists.

Every Windows beta is built with the production adapters on (D3D11, Media Foundation, WASAPI, Zoom, UVC, WGC, virtual camera). The stub core is a test configuration, not a product tier. The reference run below is the 30-minute outputs soak of October 22, 2026 on beta-2026-10-22-f96a136 (ruling and summary in docs/BACKLOG.md, "Owner gate ruling, October 22"; raw evidence in the local artifacts/qualification-1008-25, not in git): eight 1080p Zoom feeds, webcam, moving screen capture, Program + eight ISO recordings on hardware encoders, the registered OS camera at 1080p60 and an SRT stream, with zero missing frame identities on every output. beta-2026-10-23-1072a50 was not soaked. beta-2026-10-25-1f697ca had a 30-minute clean-picture soak on October 25 (no identity stamps, so no identity verdict; write-up in docs/qa/rc-beta-acceptance-2026-10-25.md): the stream, every ISO writer and the camera publisher ran clean, and Program recording missed one frame on a 15 ms render-tick lock hold (#1056).

Every operator-visible value is measured or absent; a placeholder is never shown (honest reporting). The release gate is per-frame identity of every output over the full workload (a writer-reported held frame during a source stall passes as a labelled, counted repeat, owner ruling 2026-10-23; see the outputs drill); on-screen Studio cells are a tolerance metric because Windows DWM skips frames at the mixed-refresh monitor beat for every application (scripts/qa/desktop-picture-cadence.py).

What stands between this build and a release candidate is tracked only in docs/BACKLOG.md. The non-engineering items are a code-signing certificate, the Zoom SDK redistribution answer, reference machines and a tester feedback channel (#423 to #427); the controlled-rig two-hour soak in docs/release-evidence.md has not been run.

Run the app (Windows), one command:

npm install
npm run app                 # builds the best-available native core, then builds + launches the WinUI app

npm run app auto-picks the media-core tier it can build on your machine — full (Zoom + D3D11 + RTMP + audio, needs the Zoom SDK), audio (real Windows audio, synthetic video), or stub (all synthetic) — rebuilds it only when stale, then launches and prints which tier you got. Force a rebuild with npm run app -- -Rebuild.

Other commands:

npm run test:unit           # vitest: services + Companion module tests
npm run test:show-engine    # show-engine workspace tests
npm run test:scripts        # node:test suites under scripts/
npm run build:studio        # C++ media core + native Studio shell (Windows)
npm run run:studio          # launch the native Studio shell (Windows)
npm run pack:native         # stage the WinUI shell + native core for distribution

Full Windows gate (contracts + show-engine/native/shell suites): npm run test:gate. Offline readiness report: npm run alpha:preflight.

On macOS, build the Swift shell with cd mac-shell && swift build -c release. Run its deterministic suite with COREVIDEO_SHELL_TESTS=1 .build/release/CoreVideoProShell. The macOS launch script documents the native core and SDK configuration. A successful shell build does not validate the real media adapters.

The first fully useful milestone:

1. Join Zoom.
2. See clean participant feeds and metadata.
3. Click Magic Scene.
4. Get a polished show with lower-thirds, graphics, audio leveling, and stream / local-recording / virtual-camera controls.

• Ranked work order: docs/BACKLOG.md.
• Working guide (build/run/architecture/crash class/current state): CLAUDE.md.
• GPU multiview implementation plan: docs/gpu-multiview-plan.md.
• Operator performance investigation & plan: docs/operator-performance-plan.md · present-stutter fix spec: docs/present-stutter-fix-spec.md.
• Sources redesign, preview/program direct positioning, browser sources: docs/sources-redesign-spec.md (builds on docs/capture-sources-spec.md).
• Compositor architecture (OBS/CasparCG/Natron-informed): docs/compositor-architecture-plan.md.
• Sprint-by-sprint demo roadmap: docs/roadmap/index.html (open in a browser).
• Alpha build plan & exit bar: docs/alpha-plan.md.
• Focus & competitiveness plan (cut line, demos, 90-day sequencing): docs/FOCUS_PLAN.md.
• Native production completion plan: docs/native-production-completion-plan.md.
• Product spec & positioning: COREVIDEO_PRO_PRODUCT_SPEC.md.
• WinUI shell setup: native-shell/README.md · Studio shell: studio/README.md.