Lesson 258: OBS Replay Buffer Uniform Fragments Receipt on BUILD_RECEIPT (2026)

Direct answer: Before June–October 2026 batch Whisper nights, promote replay_buffer_uniform_receipt_v1.json proving facilitators share one OBS playtest profile (fixed Replay Buffer seconds, 48 kHz project audio, ≥2 s between saves), replay_buffer_uniform_ladder_v1.json pins the sample-rate ladder, and +genpts is documented on the concat path—distinct from Lesson 200 (concat_ok on playtest_vod_triage_receipt_v1), Lesson 207 (path fork when merge fails), and Lesson 243 (highlight consent). Pair the OBS uniform fragments preflight (Guide #21).

Why this matters now (batch Whisper nights)

October 2026 facilitators save 12–20 Replay Buffer MKVs per Discord thread—per-clip Whisper works, ffmpeg -f concat fails on mixed 44100 vs 48000 Hz, non-monotonic DTS, and VFR offsets. Teams open MKV concat help at 11pm instead of pinning OBS before capture—258 files replay_buffer_uniform_receipt_v1.json so BUILD_RECEIPT replay_buffer_uniform_ok blocks promotion when fragment policy drifts. Concat decision tree picks paths after concat_ok: false; this lesson prevents the fork.

The OBS uniform fragments preflight is the ninety-second gate—258 is the BUILD_RECEIPT milestone.

Beginner path (OBS profile → ladder → receipt)

Time: ~55 minutes first batch night; ~12 minutes when profile is cloned.

Developer path (gates R1–R6)

R1 — cousin receipt crosswalk

Do not merge schemas—reference paths in cousin_receipts only.

replay_buffer_uniform_ladder_v1.json

{
  "schema": "replay_buffer_uniform_ladder_v1",
  "build_label": "october-fest-2026-playtest-rc3",
  "obs_profile": "Playtest_48k_120s",
  "replay_buffer_seconds": 120,
  "audio": {
    "project_sample_rate_hz": 48000,
    "normalize_target_hz": 48000,
    "channels": "mono",
    "allowed_rates_in_fragments": [48000]
  },
  "save_spacing_seconds_min": 2,
  "concat_policy": {
    "prefer_wav_ladder": true,
    "mkv_remux_genpts": true,
    "ffmpeg_genpts_flag": "+genpts"
  },
  "fragment_manifest_columns": ["clip_id", "wall_time_utc", "duration_s", "sample_rate_hz", "sha256"]
}

Pin under release-evidence/playtest/replay-uniform/REPLAY_BUFFER_UNIFORM_LADDER.json.

R4 — concat policy snippet (WAV ladder + genpts)

# Per-fragment normalize (R3 ladder)
ffmpeg -i "$frag" -ar 48000 -ac 1 "normalized/${base}.wav"

# MKV remux path when WAV skipped — genpts required (R4)
ffmpeg -i "input.mkv" -c copy -fflags +genpts "work/remux_${base}.mkv"

# Concat WAV list (preferred)
ffmpeg -f concat -safe 0 -i fragments.txt -c copy merged_playtest.wav

replay_buffer_uniform_receipt_v1.json

{
  "schema": "replay_buffer_uniform_receipt_v1",
  "build_label": "october-fest-2026-playtest-rc3",
  "ladder_path": "release-evidence/playtest/replay-uniform/REPLAY_BUFFER_UNIFORM_LADDER.json",
  "ladder_sha256": "cc33…",
  "obs_profile_export": "release-evidence/playtest/replay-uniform/obs_profile_Playtest_48k_120s.json",
  "facilitator_roster_sha256": "dd44…",
  "cousin_receipts": {
    "vod_triage_concat_ok": "release-evidence/playtest/vod/PLAYTEST_VOD_TRIAGE_RECEIPT.json",
    "mkv_gap_reencode": "release-evidence/playtest/mkv-gap/MKV_GAP_REENCODE_RECEIPT.json",
    "clip_consent": "release-evidence/playtest/consent/PLAYTEST_CLIP_CONSENT_RECEIPT.json"
  },
  "gates": {
    "R1_readme_profile": "pass",
    "R2_buffer_length": "pass",
    "R3_sample_rate_48k": "pass",
    "R4_genpts_policy": "pass",
    "R5_ladder_committed": "pass",
    "R6_build_receipt": "pass"
  },
  "replay_buffer_uniform_ok": true,
  "whisper_batch_prerequisite_met": true
}

Pin under release-evidence/playtest/replay-uniform/REPLAY_BUFFER_UNIFORM_RECEIPT.json.

BUILD_RECEIPT row (R6)

ALTER TABLE release_publish_gate ADD COLUMN IF NOT EXISTS
  replay_buffer_uniform_blocked BOOLEAN NOT NULL DEFAULT false;

Thursday row review — Replay uniform line: profile + 48 kHz Y/N.

Key takeaways

1. Uniform at capture beats ffmpeg heroics at midnight (R1–R3).
2. +genpts is policy, not optional trivia (R4).
3. Lesson 200 — post-merge concat_ok; 258 — pre-night profile.
4. MKV concat help — reactive triage cousin.
5. Concat decision tree — path after concat_ok: false.
6. Lesson 243 — legal consent column.
7. Cousin: Lesson 259 Bevy menu flush—parallel engine fest column.
8. Lesson 257 — parallel Unity strip column.
9. October capstone 265 wires 254–264 including this row.
10. Top 20 receipts hub — replay uniform row in fest→public map.

Common mistakes

• Fixing concat only in ffmpeg while OBS profile still 44100 Hz (R3).
• Mixed Replay Buffer 30 s vs 120 s across facilitators (R2).
• Filing 200 while uniform receipt RED—merge vs capture are different gates.
• Merging replay_buffer_uniform_receipt into playtest_vod_triage_receipt.
• Skipping +genpts on MKV path then blaming Whisper (R4).

Troubleshooting

Mini exercise (45 minutes)

1. Export two MKVs from different OBS profiles—confirm sample-rate mismatch in ffprobe.
2. Clone Playtest_48k_120s—re-export; confirm 48000 Hz only.
3. Commit ladder JSON; run R5 hash pin.
4. File receipt; BUILD_RECEIPT GREEN.

Continuity — October–Q4 2026 fest ops truth (254–265)

Previous: Lesson 257 — Unity Addressables jq audit
Next: Lesson 260 — Unity UGS/OTLP diligence (queued)

FAQ

Same as Lesson 200?
200 = concat_ok after merge; 258 = uniform OBS policy before capture nights.

Same as Help #5?
Help is reactive ffmpeg; 258 is BUILD_RECEIPT prevention + ladder artifact.

Still need per-clip Whisper?
Yes when decision tree picks per_clip_local—uniform policy reduces how often you fork.

WAV ladder vs MKV +genpts?
Ladder prefers WAV concat; MKV remux with +genpts is documented fallback (R4).

Batch Whisper nights fail when every facilitator uses a different Replay Buffer profile—pin 48 kHz, fixed seconds, genpts policy, ladder JSON, then replay_buffer_uniform_ok before triage promotion.