# Erin Audio Pack — v2 (corrected)

**For:** Kelly Mortazavi, Avina Wellness
**From:** Fabian
**Date:** 03/05/2026

---

## v1 -> v2 corrections

This is the v2 build of the Erin audio set, replacing the v1 build (folder `Keli_Erin_Audio_Pack_2026-05-02`, since removed). Three errors in v1 are corrected here:

1. **Speaker labels were swapped.** v1 used a word-count heuristic that assumed the dominant speaker in each file was Kelly. In this case set the dominant speaker is Erin (the client in crisis); Kelly's interventions are short. Every quote attributed to Kelly in v1 below the threshold of independent verification was actually Erin, and vice versa. Speaker attribution has been re-derived by reading the actual content (case-manager language, addresses of Erin by name, vs client-in-crisis language).
2. **Per-line metadata in transcript bodies removed.** v1 included timestamps, engine confidence scores and AGREE/DISPUTED tags inline with each line, which made the documents read as machine output. v2 transcripts are clean speaker turns: `Kelly: ...` / `Erin: ...` paragraphs only.
3. **File 02 (2-minute fragment) is now flagged as silence.** v1 rendered engine hallucination ("This video is a work of fiction" YouTube boilerplate) as if it were real speech. v2 replaces that file's body with a methodology note explaining no transcript can be produced. The audio file remains in `01_AUDIO/` for completeness.

## What's in this folder

| Folder | Contents |
|---|---|
| `01_AUDIO/` | The six original m4a audio files. Unchanged. |
| `02_TRANSCRIPTS_FINAL/` | Six rebuilt transcripts (DOCX + PDF). v2 build. |
| `03_TRANSCRIPTS_PLAUD/` | Your six original PLAUD transcripts as captured. Unchanged. |
| `04_REFERENCE/` | Six reference documents, all rebuilt with corrected attribution. |
| `05_INCIDENT_REPORT/` | Your incident report PDF (unchanged) and the rebuilt alternative incident report (v2). |
| `06_FOR_FAMILY/` | Position document for Carol (rebuilt). |
| `07_SYSTEM_BUILD/` | The Avina Methodology (rebuilt). |
| `08_EMAIL_TO_KELLY/` | Cover note for this v2 set. |
| `09_INTERACTIVE/` | The Methodology Visual Canvas, the Client Portal v1, and the Client Portal v2 (interactive HTML). |

## Reference documents (all v2)

- **AVN-CLN-001.2 — Findings.** Methodology document explaining the v2 build, including the v1 -> v2 correction note.
- **AVN-CLN-CMP-001.2 — PLAUD vs Dual-Engine Comparison.** Updated for v2.
- **AVN-CLN-RPT-001.2 — Per-File Summary Report.** One section per recording, confirming v1 -> v2 correction per file.
- **AVN-CLN-IDX-001.2 — Topic Index.** Themes across the six recordings, attributed correctly.
- **AVN-CLN-CHR-001.2 — Consolidated Chronological Transcript.** All six files in PLAUD recording order, with corrected attribution.
- **AVN-CLN-QTR-001.2 — Quote Register.** Flat list of every notable quote, attributed correctly.

## A note on the new transcripts

The dual-engine final transcripts use two colours to make speaker attribution easy to scan:
- **Kelly** — navy
- **Erin** — red
- Other speakers — grey

There are no per-line timestamps, confidence scores, or AGREE/DISPUTED tags in the body. Methodology and engine details are described in the document header. If a future audit needs per-utterance data, the underlying Gladia and Deepgram JSON files are retained at the source location.

## What this folder is and is not

This is the source-document version of the Erin audio set. It contains the recordings, the rebuilt transcripts, reference tools, your own original incident report, and a small set of internal Avina documents (alternative incident report, methodology, position document for Carol, two interactive client portal prototypes). There are no operational instructions for what to do with any of it. The decisions about what to do with the material are yours.

If you want any deeper analytical or operational documents at any point, ask and they can be produced.