{
  "schema": "ghostmesh-video-receipt-v1",
  "createdAt": "2026-10-17T12:29:29+00:00",
  "source": {
    "url": "https://www.youtube.com/watch?v=Z8lf5oWKfj0",
    "videoId": "Z8lf5oWKfj0",
    "title": "Matt Pocock: The Most Valuable Skill to Learn in the Age of AI",
    "channel": "Nate Herk | AI Automation",
    "uploadDate": "20261016",
    "durationSeconds": 2690,
    "captionType": "YouTube automatic English captions",
    "captionTrack": "en-orig",
    "creatorChapters": [
      {
        "startSeconds": 0,
        "title": "Intro"
      },
      {
        "startSeconds": 110,
        "title": "When Coding Agents Clicked"
      },
      {
        "startSeconds": 264,
        "title": "Tell Agents What You Care About"
      },
      {
        "startSeconds": 579,
        "title": "Where the Opportunities Are"
      },
      {
        "startSeconds": 895,
        "title": "Unknown Unknowns & Grill Me"
      },
      {
        "startSeconds": 1183,
        "title": "How to Stand Out"
      },
      {
        "startSeconds": 1563,
        "title": "The Spec-Driven Development Failure"
      },
      {
        "startSeconds": 1848,
        "title": "Models, Harnesses & Effort Levels"
      },
      {
        "startSeconds": 2127,
        "title": "How to Get Started"
      },
      {
        "startSeconds": 2491,
        "title": "What's Next for Matt"
      }
    ]
  },
  "reasonPack": {
    "schema": "ghostmesh-reasoning-pack-v1",
    "verification": "pass",
    "formatSelection": "136+251-6 (inspected 720p H.264 video plus Opus audio; normalized to MP4/M4A)",
    "manifestSha256": "06a46ae8e66561e9b53af9ae79ec22820f2dab96012b8a7bad769dbd02f202a9",
    "members": {
      "video.mp4": {
        "sha256": "60c654fe0368e09db9475acef1c484608393ba9ca5ddb7e39c1c855c5fa7cb87",
        "bytes": 369062828
      },
      "audio.m4a": {
        "sha256": "177613de619f9b07c08d894648055ec1230ca0cf67662f83cb09d98cc36a9a19",
        "bytes": 41804346
      },
      "transcript.txt": {
        "sha256": "f4ff5be366d1db4b60570eb29dbf0ee577d441abd765cd0784c608afb1ee6771",
        "bytes": 47021
      },
      "transcript-status.json": {
        "sha256": "de57500cafe4856dd6f43fa1cd0216e975b49dbbbcbe09eafb30ee5318cd7565",
        "bytes": 284
      },
      "source.json": {
        "sha256": "b05383fab4fc54acdc307418d1e47de27872a8dee9d5d84b8962b90cec9aee82",
        "bytes": 11655839
      },
      "ffprobe.json": {
        "sha256": "1fd1b7165a7a70b0d30e97ae672e0e17ff25282744b9feca0c33c25158045dcd",
        "bytes": 4613
      },
      "captions.vtt": {
        "sha256": "c353872fd417831d29c4d0ea1f7c923795d671e6e88e2c70ad55e4042198fbc3",
        "bytes": 447859
      }
    },
    "publicationBoundary": "The public site distributes only original commentary and a sanitized claim-level evidence receipt. Original video, audio, full transcript, and source captions remain in a private, hash-verifiable ReasonPack."
  },
  "derivation": {
    "captionAcquisition": "The built-in ReasonPack caption command did not retain a track and began local Whisper fallback. That run was stopped; a separate yt-dlp en-orig VTT fetch succeeded. The source VTT and transcript were inserted into the completed media pack, its manifest and sums were regenerated, and ReasonPack verify passed.",
    "transcript.txt": "Incremental lines in the ReasonPack VTT (cue lines with word-level timestamp tags), HTML entities decoded, grouped by 30-second cue start; no wording correction or speaker attribution.",
    "captions.vtt": "Byte-for-byte copy of ReasonPack captions.vtt.",
    "verification": "ReasonPack closed-manifest SHA-256 verification, followed by SHA-256 calculation of public files."
  },
  "editorial": {
    "verdict": "Take the core lesson seriously; watch selected passages, not necessarily the full 44:50 interview. Start with a reversible, checkable task. Seek a separate deployment and security playbook before broader access.",
    "thesis": "Good agent work depends on closing the loop from intent to inspectable behavior, not merely giving longer instructions.",
    "recommendedSegments": [
      {
        "startSeconds": 1597,
        "endSeconds": 1760,
        "why": "First-person failure story about spec-driven work in a video editor."
      },
      {
        "startSeconds": 1923,
        "endSeconds": 2050,
        "why": "First-person account of improving the agent environment with app interaction and screenshots."
      },
      {
        "startSeconds": 2161,
        "endSeconds": 2487,
        "why": "Start small, observe the agent, expand trust with evidence, and prefer checkable tasks."
      }
    ],
    "claimMap": [
      {
        "id": "communication",
        "sourceStartSeconds": 264,
        "kind": "speaker-opinion",
        "finding": "Pocock says agents need explicit priorities and a well-maintained codebase.",
        "limit": "No comparative study in the interview."
      },
      {
        "id": "primary-sources",
        "sourceStartSeconds": 439,
        "kind": "speaker-practice",
        "finding": "He says he uses agent teaching as an introduction before reading primary material.",
        "limit": "The practice is described, not independently audited."
      },
      {
        "id": "spec-failure",
        "sourceStartSeconds": 1597,
        "kind": "first-person-case",
        "finding": "He reports his video-editor code and agent results worsened while centering a large spec.",
        "limit": "One project, no shared logs or controlled comparison."
      },
      {
        "id": "environment-fix",
        "sourceStartSeconds": 1991,
        "kind": "first-person-case",
        "finding": "He reports better results after agents could run and inspect his application.",
        "limit": "No before/after quality metric or artifacts in the interview."
      },
      {
        "id": "productivity",
        "sourceStartSeconds": 1850,
        "kind": "first-person-claim",
        "finding": "He describes multiple agents and hundreds of commits per day.",
        "limit": "Commit count is not a quality or value measure; not independently checked here."
      },
      {
        "id": "trust-ladder",
        "sourceStartSeconds": 2161,
        "kind": "speaker-advice",
        "finding": "He recommends supervised, small tasks before widening agent scope.",
        "limit": "Permission, approval, and rollback controls are GhostMesh recommendations."
      },
      {
        "id": "sales-advice",
        "sourceStartSeconds": 1255,
        "kind": "speaker-advice",
        "finding": "He suggests a client-specific prototype and acknowledges limited experience competing without a large audience.",
        "limit": "Not a validated sales strategy; an aside about billing should not be treated as guidance."
      }
    ],
    "externalPrimarySources": [
      {
        "url": "https://github.com/mattpocock/skills",
        "role": "Author-owned public skills repository; existence check only."
      },
      {
        "url": "https://ai.engineer/talks/8rABwKRsec4-specifications-are-the-new-code",
        "role": "Sean Grove talk that Pocock invokes; its position on specs is not a measured comparison of his project."
      }
    ],
    "limits": [
      "YouTube automatic captions may mishear names, model versions, and technical terms.",
      "No speaker diarization or manual transcript correction.",
      "No independent audit of repositories, sessions, costs, commit history, client outcomes, or security controls.",
      "Editorial judgments are not empirical effect sizes."
    ]
  }
}