Rights & provenance
Content ID workflow
Generative audio only earns trust if you can prove where it came from. Eleven steps run on every render, and the last one is public.
- 01hr_series[]
Capture
Wearable streams heart rate at 1 Hz. Only physiological channels are read; the stream is discarded after the session summary is computed.
- 02rx_…
Prescribe
Tanaka HRmax and Karvonen reserve produce the zone band; the goal preset produces six phases with per-phase heart-rate targets.
- 03tempo_bpm
Anchor tempo
Each phase target folds into a musical BPM inside 70–190, so the beat sits at a whole-number relationship to the heartbeat.
- 04render_job
Compose
The generative model receives the tempo map, mood token and phase energies — never identity, never raw health records.
- 05disclosure_tag
Disclose
An AI-generation disclosure is written into container metadata and shown in the player, per FTC 16 CFR §255 and EU AI Act Article 50.
- 06watermark_payload
Watermark
An inaudible payload carrying prescription and session ids is embedded, surviving transcode and re-encode.
- 07audio_sha256
Hash
SHA-256 over the rendered audio object binds the file bit-for-bit to the record.
- 08row_hash
Seal
One append-only row is written, including the hash of the previous row. Updates and deletes are refused at the database level.
- 09A-VO2-…
Fingerprint
The render is fingerprinted for reference delivery, with the asset id reserved at generation time.
- 10content_id_asset
Deliver reference
Reference audio plus ownership metadata goes to the rights registry so third-party uploads match back to the originating athlete session.
- 11GET /api/public/v1/audit
Publish or verify
Anyone can query the public audit endpoint by prescription id and re-verify the chain without seeing a single identifying field.
Want to see it end to end? Generate a session, seal it, then look it up in the explorer.