# FlowPick tool samples

These assets are generated test patterns, sine waves and synthetic speech, created
for FlowPick on 2026-10-19 and expanded on 2026-10-28. They contain no third-party
recordings, user speech, account data or signed URLs.
You may use and redistribute them for testing without attribution (CC0-1.0).

- `flowpick-test.mp4`: 4 seconds, 160 × 90, H.264, 25 fps; mono AAC at 44.1 kHz.
- `flowpick-test.m3u8`: VOD, 2 MPEG-TS segments of 2 seconds each, no encryption.
- `flowpick-test.mpd`: static DASH, one video and one audio representation.
- `flowpick-tone.wav`: 2 seconds, 440 Hz, mono signed 16-bit PCM at 44.1 kHz;
  176,444 bytes including the 44-byte WAV header.
- `flowpick-test.webm`: the short pattern encoded as VP8/Opus, 160 × 90.
- `flowpick-animation.webp`: 20 animated frames over 2 seconds, 160 × 90, no audio.
- `flowpick-speech.wav` and `flowpick-speech.mp4`: synthetic English speech,
  generated with FFmpeg/libflite's built-in `slt` voice and our own two sentences.
  The WAV is mono at 16 kHz. `flowpick-speech.txt` is the reference transcript.
  Duration is measured in `verification.json`, not assumed from the video pattern.
- `flowpick-noisy-tone.wav`: 2-second tone plus deterministic noise (seed 440),
  a control signal rather than a noisy-speech quality benchmark.
- `flowpick-invalid.m3u8`: plain text without EXTM3U, for an invalid-input check.
- `flowpick-dash-sample.zip`: the MPD and its complete relative dependency set.
- `verified-hls.mp4`: output generated by the website's actual TS remux helper,
  not by FFmpeg. It is a code-level reference, not an export from a browser UI test.

The TS and M4S files in this directory are dependencies of the manifests. Keep
their relative filenames intact when copying or deploying the samples.

These inputs exercise short unencrypted workflows. They do not validate DRM,
authenticated sources, live streams, multi-period DASH, large-file performance,
speech recognition accuracy, source-site compatibility, or browser save permissions.

## Verification record

Run `npm run check:tool-samples` from the repository root with Node.js and ffprobe.
The runner calls the website's HLS parser, media-info parser, WebM inspector and
TS remux helper. It also verifies deliberately invalid inputs and fixture properties.
ffprobe independently inspects the helper's MP4 output. Assertions must pass before
the JSON report is published. `verification.json` records actual runtime versions,
timestamp, per-check scope and SHA-256 hashes of the downloadable files.

The report is mirrored to `shared/data/toolVerification.json` for prerendering.
After regenerating fixtures or changing the tested media functions, rerun verification
before building. Browser codec availability, UI completion, AI inference, word error
rate, audio quality and processing speed are **not** certified by these Node.js checks.
No fabricated browser/device compatibility results or performance measurements are used.

For a browser reproduction, record browser/version, OS, settings, selected input hash,
output metadata, playback and whether the file was actually saved. WAV trimming at
0.500–1.500 s is a one-second expectation, not a claimed measured export. AI speech
recipes include a transcript for comparison but claim no recognition score. The tone
and noisy-tone controls do not establish vocal-separation or speech-denoising quality.

Regenerate with `scripts/generate-tool-samples.ps1` from the repository root.
It requires local FFmpeg (libflite, libvpx, libopus, libwebp_anim) and Python;
neither is needed by visitors using the tools. Then run `npm run check:tool-samples`.
Encoded bytes can vary between FFmpeg versions.