**Help page (future link):** `[HELP_SITE_URL]/how-it-works`

## The Core Idea
Onda Replay separates two things that most tools conflate:
1. **The recording** — audio and transcript. These are content. They belong to the creator.
2. **The annotations** — phrases, tags, markers. These are personal. They belong to the listener.

The `.onda` bundle delivers the content. Each listener builds their own annotations.
User Roles

User A — The Producer (Professor, Presenter, Host)
- Records the session in Onda Replay
- Transcribes once (~$0.006/min via Whisper — roughly **$30/semester** for daily lectures, or **free** with more manual effort)
- Labels the recording and adds notes
- Shares the `.onda` bundle: **audio + transcript + metadata**
- Done. Their work is complete.

User B — The Active Listener (Student who attended)
- Was present during the recording
- Taps **|> Mark this moment** during the live session whenever something important is said
- After the session, transcribes their own copy — markers become anchor points for Whisper phrase generation
- Phrases appear in their **Phrases drawer** as a personal study index
- Can share their own `.onda` bundle — **without** their markers or phrases (by design)

 User C — The Remote Listener (Student who did not attend)
- Receives the `.onda` bundle from User A or B
- Imports it — gets audio + transcript immediately, no transcription cost
- Two paths to build their own annotations:
  - **Listen path:** Play recording in Listening Room → tap **|> Mark this moment** at key moments → transcribe to generate phrases
  - **Read path:** Open transcript → listen while reading → long-press key phrases → add to Phrases drawer manually
- Their phrases reflect *their* understanding — which may differ entirely from User B's

## What Travels in the .onda Bundle — and What Doesn't

| Item | In `.onda` bundle? | Why |
| Audio file | ✓ Yes | It's the content |
| Transcript text | ✓ Yes (if transcribed) | It's the content |
| Label & notes | ✓ Yes | Metadata about the recording |
| Recording timestamp | ✓ Yes | Provenance |
| Tap markers | ✗ No | Personal annotation |
| Phrases | ✗ No | Personal annotation |
| Tags | ✗ No | Personal annotation |

Phrases are not prescribable.

What matters to one listener may be irrelevant to another. The bundle is a delivery format, not a collaboration format.

The Productive Friction Principle

The manual path (read transcript → highlight → add phrases) requires more effort than the automated path (mark during recording → transcribe → phrases auto-generated).

That friction is intentional. The cognitive work of selecting what matters *is the learning*. Passive annotation is not the goal.

Cost Model

| Scenario | Transcription cost | Who pays |
|---|---|---|
| Professor transcribes, shares with 30 students | Once | Professor (or department) |
| Each student transcribes independently | 30× | Each student |
| Student imports bundle with transcript already included | Zero | None |

The professor-pays model reframes Onda Replay as an institutional tool rather than a per-user subscription. One person bears the cost; everyone benefits.

## Design Decisions Captured Here

- `.onda` bundle intentionally excludes personal annotations — this is not a defect
- Phrases drawer is always owned by the individual listener
- Transcription is a one-time cost per recording, not per listener
- The Listening Room "Mark this moment" button is most useful *before* transcription exists; the text-highlight path is most useful *after*
- The air gap between app and AI is deliberate — transcript leaves the app, user interacts with AI externally, phrases come back via manual entry

---

*Draft — last reviewed 2026-03-13*
*Help page link: placeholder — update when `[HELP_SITE_URL]` is confirmed*