Skip to main content

🎥 Video Walkthrough Scripts

Short team-facing scripts aligned to the current non-GPU architecture.

:::note Recording status These are scripts, not proof that a video was recorded or published. Verify the UI again immediately before recording because screen copy changes faster than architecture. :::

Architecture in five minutes​

0:00 — Context​

“AceSense has a Flutter player/coach app, public landing site, admin panel, private docs and launchpad, and a public REST/MCP API. Firebase Auth, Firestore, Storage, Hosting, and Cloud Functions tie those surfaces together.”

1:00 — Upload path​

“The Flutter app validates one selected video and calls requestUploadPath in europe-west1. The server checks consent, entitlement, quota, and metadata, creates a pending Firestore job, and returns owner-scoped Storage paths. The client uploads one file; client-side physical chunking is not shipped.”

2:15 — Async processing boundary​

“A Storage finalize event validates the object and enqueues dispatchGpuJob. That Firebase Task Queue worker owns retries and provider dispatch. The GPU service is an external boundary for this walkthrough; its internals live in its own repository.”

3:00 — Completion​

“When a combined result appears, Functions atomically claim the terminal job state, release quota once, generate derived insights/report artifacts, and write URLs and summary data. The Flutter app is subscribed to the job, so the completed analysis appears without a manual refresh.”

4:00 — Operations and access​

“Admin actions go through requireAdmin; owners cannot forge job status. Docs and launchpad are not ordinary public SPAs: all requests pass through a Function that verifies an admin session cookie before serving bytes.”

Frontend setup in ten minutes​

  1. Show mise.toml pinning Flutter 3.41.9.
  2. Run flutter pub get and flutter doctor.
  3. Run flutter analyze --no-fatal-infos and one focused test.
  4. Launch Chrome with flutter run -d chrome.
  5. Tour lib/features, lib/services, lib/routing, and lib/ds.
  6. End with flutter test --coverage and the release build command.

Admin incident drill​

Use a test/emulator job or an explicitly approved production incident:

  1. Find the job and read its status, timestamps, paths, and error fields.
  2. Check the owning Function logs.
  3. Distinguish abandoned upload, dispatch failure, and result-finalization failure.
  4. Use the guarded retry/status callable exposed by Admin when appropriate.
  5. Verify the resulting audit row and user-visible state.

Never demonstrate a direct Firestore status edit as the normal recovery path.