Lesson 273: Audacity Playtest VOD Loudness Spot Whisper Receipt on BUILD_RECEIPT (2026)

Direct answer: After merged playtest WAV exists with concat_ok: true but before Whisper batch scripts run, promote playtest_vod_loudness_spot_receipt_v1.json with A1–A6 gates—ffprobe duration + audio stream, Audacity −14 LUFS ±2 on a speech minute, peak < −1 dBFS, fail-closed batch jq, and BUILD_RECEIPT loudness_spot_ok: true. Pair Audacity loudness preflight (Guide #35), Audacity loudness evening tutorial, and cousin Lesson 266 (GitHub tags run after transcript exists).

Why this matters now (November Whisper batch nights)

November 2026 facilitators merge twelve OBS fragments, launch Whisper, and get [inaudible] or three-word hallucinations while Discord claims an hour of speech. concat_ok is true—merge integrity passed—but the merged WAV is clipped, silent on the wrong track, or menu-music-only on the spot minute. Teams blame ASR and open empty GitHub tag issues—scenario E in Whisper tags strategy.

Lesson 272 wired Ren'Py upload wall-clock; 273 wires acoustic spot before Whisper so Thursday row review diffs loudness_spot_ok beside concat_ok—not founder memory of “sounds fine in VLC.”

Promotion order on BUILD_RECEIPT: concat_ok → loudness_spot_ok → Whisper batch → whisper_issue_tag_ok.

Beginner path (merged WAV → spot → receipt)

Time: ~45 minutes first wiring; ~8 minutes when Guide #35 preflight template exists.

Developer path (gates A1–A6)

Align with Guide #35—guide = ninety-second gate; 273 = BUILD_RECEIPT milestone with jq promotion.

Cousin receipt crosswalk

Do not merge JSON schemas—reference paths in cousin_receipts only.

A1 — Merged WAV + concat cousin

WAV="playtest-vod/merged/session_2026-11-12_merged.wav"
test -f "$WAV"
jq -e '.concat_ok == true' playtest-vod/receipts/CONCAT_RECEIPT.json

Red flag: Whisper batch on per-clip MKVs when evening concat was skipped—fix MKV concat help first.

A2 — ffprobe duration and stream

ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "$WAV"
ffprobe -v error -select_streams a:0 -show_entries stream=codec_name,sample_rate,channels -of csv=p=0 "$WAV"

Pass example: 3724.512000 and pcm_s16le,48000,2. Fail: duration N/A or < 60 s when session should be longer.

A3–A4 — Audacity spot minute

1. Open merged WAV in Audacity 3.x.
2. Select a representative minute with player speech (not menu-only).
3. Analyze → Loudness and Peak… — record integrated LUFS (target −14 ±2).
4. Confirm true peak below −1 dBFS on same selection.
5. Screenshot → playtest-vod/loudness-spot/lufs_spot_<session_id>.png.

Pair 14 free LUFS meter tools when Audacity meter unavailable.

A5 — Fail-closed batch guard

#!/usr/bin/env bash
set -euo pipefail
REC="${1:-playtest-vod/loudness-spot/PLAYTEST_VOD_LOUDNESS_SPOT_RECEIPT.json}"

jq -e '.loudness_spot_ok == true' "$REC" || {
  echo "Whisper batch blocked: loudness_spot_ok false (A5)" >&2
  exit 1
}

Wire into nightly batch script before faster-whisper or local Whisper CLI—pair January batch strategy W3 loudness requirement and February same-day Whisper W3 cousin.

A6 — playtest_vod_loudness_spot_receipt_v1.json

{
  "schema": "playtest_vod_loudness_spot_receipt_v1",
  "build_label": "november-closeout-2026-rc2",
  "session_id": "session_2026-11-12",
  "merged_wav_path": "playtest-vod/merged/session_2026-11-12_merged.wav",
  "ffprobe_duration_sec": 3724.512,
  "ffprobe_audio_stream": "pcm_s16le,48000,2",
  "spot_minute_start_sec": 840,
  "integrated_lufs": -13.8,
  "true_peak_dbtp": -2.1,
  "lufs_screenshot": "playtest-vod/loudness-spot/lufs_spot_session_2026-11-12.png",
  "cousin_receipts": {
    "concat_ok": "playtest-vod/receipts/CONCAT_RECEIPT.json",
    "vo_consent": "release-evidence/vo/VO_CONSENT_METADATA_RECEIPT.json"
  },
  "gates": {
    "A1_merged_wav": "pass",
    "A2_ffprobe": "pass",
    "A3_integrated_lufs": "pass",
    "A4_peak_headroom": "pass",
    "A5_fail_closed": "pass",
    "A6_receipt": "pass"
  },
  "loudness_spot_ok": true,
  "whisper_batch_allowed": true
}

Pin under playtest-vod/loudness-spot/PLAYTEST_VOD_LOUDNESS_SPOT_RECEIPT.json.

BUILD_RECEIPT promotion jq

jq -e '
  .concat_ok == true and
  .loudness_spot_ok == true
' BUILD_RECEIPT.json

jq -e '
  .schema == "playtest_vod_loudness_spot_receipt_v1" and
  .loudness_spot_ok == true and
  (.integrated_lufs >= -16 and .integrated_lufs <= -12)
' playtest-vod/loudness-spot/PLAYTEST_VOD_LOUDNESS_SPOT_RECEIPT.json

Append BUILD_RECEIPT row:

{
  "loudness_spot_ok": true,
  "playtest_vod_loudness_spot_receipt": "playtest-vod/loudness-spot/PLAYTEST_VOD_LOUDNESS_SPOT_RECEIPT.json",
  "whisper_batch_allowed": true
}

Rule: Do not set whisper_issue_tag_ok until Whisper produces non-empty transcripts—273 gates batch start, not GitHub.

Proof table (facilitator audit)

Key takeaways

1. concat_ok then loudness—never reverse BUILD_RECEIPT order.
2. ffprobe proves stream exists; Audacity proves speech is intelligible.
3. −14 LUFS ±2 is the November batch spot target for merged playtest voice.
4. Evening tutorial owns blocks A–D; 273 owns BUILD_RECEIPT promotion.
5. Lesson 266 runs after transcripts exist—fix audio here first.
6. VO consent Lesson 250 cousin—legal metadata ≠ loudness spot.

Common mistakes

• Running Whisper when concat_ok is false—fix merge first.
• Spot-checking game-music-only minute—A3 reads wrong lane.
• Ignoring ffprobe N/A duration—A2 fail, not ASR bug.
• Boosting Discord capture to “hear better”—A4 clip kills transcripts.
• Filing GitHub issues from empty Whisper output—stop at 273, not 266.
• Merging loudness receipt into VO consent schema—separate JSON blobs.

Troubleshooting

Mini exercise (45 minutes)

1. Merge two OBS fragments with concat_ok: true on receipt.
2. Run A2 ffprobe; log duration + stream in scratch notes.
3. Audacity spot minute—capture LUFS screenshot (A3–A4).
4. File playtest_vod_loudness_spot_receipt_v1.json.
5. Run A5 batch guard with receipt removed—confirm exit 1.
6. Promote BUILD_RECEIPT loudness_spot_ok for Thursday review.

Continuity — November 2026 fest close-out + playtest triage (266–276)

Previous: Lesson 272 — Ren'Py SteamPipe wall-clock receipt
Next: Lesson 274 — Unreal OTLP corporate proxy diligence receipt (forward)

FAQ

Same as Guide #35?
Guide = ninety-second A1–A6 preflight; 273 = BUILD_RECEIPT loudness_spot_ok column + batch jq promotion.

Same as loudness evening blog?
Blog owns blocks A–D worksheet; 273 owns receipt culture on BUILD_RECEIPT.

Can we skip spot if per-clip Whisper works?
Batch merged path still needs A3–A4—per-clip does not excuse silent merge track.

February live week same-day triage?
Pair February same-day Whisper W3—same acoustic gate before P0 transcripts.

Does loudness_spot_ok replace concat_ok?
No—both required; different schemas and gates.

Related GamineAI reads

• Audacity loudness preflight (Guide #35)
• Audacity loudness spot evening tutorial
• Local Whisper VOD pipeline
• VO consent preflight
• Lesson 266 — Whisper issue tag receipt

Sign-off: Prove duration with ffprobe, prove speech with Audacity LUFS spot, promote loudness_spot_ok on BUILD_RECEIPT, then run Whisper—not the other way around.