# Manual verification: server engine (M5, D-013)

The dev environment has no Python package manager and no browser, so — like
`tests/offline.md` and `tests/live.md` — the real-inference path needs a
manual run on a machine with Python ≥ 3.9. Everything protocol- and
HTTP-level is already covered by `bun test` (`tests/transcribe.test.ts`
exercises the real `Bun.spawn` supervisor against
`tests/fixtures/fake_sidecar.py`, plus the degradation paths against the real
`server-py/sidecar.py`).

## 1. Install the sidecar dependencies

```sh
python3 -m pip install -r server-py/requirements.txt
```

## 2. Opt-in automated pass (no browser needed)

```sh
FW_INTEGRATION=1 bun test tests/transcribe.test.ts
```

Runs the real sidecar end to end (spawn → ready handshake → `tiny.en` load →
transcribe → done), then the first 60 s of `tests/audio_files/en_alien.opus`
(needs `ffmpeg`) in both decoding modes (D-047): `file` (faster-whisper's
`BatchedInferencePipeline` + Silero VAD, what `POST /api/transcribe` uses) and
`live` (sequential, what server-live passes use). First run downloads ~75 MB
of weights from the Hub. Point `SIDECAR_PYTHON` at a virtualenv's python if
faster-whisper isn't installed for `python3` itself.

Without faster-whisper at all, `bun test tests/transcribe.test.ts` still runs
the real `server-py/sidecar.py` against a stub (`tests/fixtures/stub_fw`) to
check the mode routing, the one-model cache and cancel between segments.
With `FW_INTEGRATION=1` it also pushes the whole 15-min `en_alien` clip
(`tiny.en`) through `POST /api/transcribe` — heartbeat lines before the first
burst, segments in bursts — and cancels a batched run after its first burst.

- [ ] The `real faster-whisper sidecar` describe block passes.

## 3. In-browser run (both engines, same audio)

1. `bun run dev`, open the app, register/sign in.
2. `GET /api/transcribe/status` (with the session cookie) returns
   `{"available": true}` — e.g. check the Network tab.
3. Pick `Whisper Tiny (English)`; transcribe a short clean English clip
   (e.g. the JFK "ask not…" sample) **on device**. Note the transcript.
4. Switch the engine toggle to **Server** (visible only while signed in;
   requires the M5 UI wiring — until that lands, drive it manually:
   `curl -X POST --data-binary @audio.f32 -b "sid=…" \
    "http://localhost:3000/api/transcribe?model=onnx-community/whisper-tiny.en"`
   where `audio.f32` is raw float32 mono 16 kHz PCM).
5. Transcribe the same clip through the server engine.

- [ ] Segments stream in as JSON lines (not one blob at the end). With the
      batched pipeline (D-047) they arrive in bursts — one burst per 8
      speech chunks (≈ up to 4 min of audio) — with blank heartbeat lines
      every 5 s in between; on a clip under ~30 s of speech, everything
      arrives as a single segment at the end. Each segment covers one VAD
      speech chunk (up to 30 s), not one sentence — expected (D-047).
- [ ] Cancelling a long server transcription (abort / page refresh) stops
      the sidecar within one burst; the next request then starts normally.
- [ ] Both transcripts agree in content (minor punctuation/casing drift
      between Transformers.js and faster-whisper is expected).
- [ ] Signed out, the toggle is gone and `POST /api/transcribe` answers 401.
- [ ] Second request on the same model is visibly faster (no reload —
      warm-up strategy).
- [ ] `AUTH_ENABLED=false bun run dev`: `/api/transcribe*` answers 404 and
      the anonymous browser flow is unaffected.

When this passes, check off the M5 integration item in `WorkBreakdown.md`.