{
  "schemaVersion": "1",
  "revision": "22ce5d34dbcb5a78",
  "id": "factory-reasoning-and-video-worked-example",
  "slug": "reasoning-and-video-worked-example",
  "title": "Independent reasoning and video evidence — worked example",
  "summary": "Detailed worked example with concrete prompts, observable outcomes, and the limits of the example.",
  "category": "Verification",
  "tags": [
    "reasoning-and-video",
    "examples"
  ],
  "sources": [
    {
      "label": "Independent reasoning and video evidence — worked example",
      "url": "https://shiny-rosette-m6zy.here.now/replay.html"
    }
  ],
  "sourceUnits": [
    "08-oracle-video/replay.html#Before the session: the only beat a human owns",
    "08-oracle-video/replay.html#The session, beat by beat",
    "08-oracle-video/replay.html#Where it landed: the shard prompt, verbatim"
  ],
  "collection": "factory",
  "categoryId": "verification",
  "wordCount": 1048,
  "url": "/topics/factory-reasoning-and-video-worked-example/",
  "markdownUrl": "/markdown/factory-reasoning-and-video-worked-example.md?raw=1&v=22ce5d34dbcb5a78",
  "jsonUrl": "/api/v1/documents/factory-reasoning-and-video-worked-example.json",
  "headings": [
    {
      "id": "before-the-session-the-only-beat-a-human-owns",
      "text": "Before the session: the only beat a human owns",
      "level": 2
    },
    {
      "id": "the-session-beat-by-beat",
      "text": "The session, beat by beat",
      "level": 2
    },
    {
      "id": "where-it-landed-the-shard-prompt-verbatim",
      "text": "Where it landed: the shard prompt, verbatim",
      "level": 2
    }
  ],
  "markdown": "# Independent reasoning and video evidence — worked example\n\nDetailed worked example with concrete prompts, observable outcomes, and the limits of the example.\n\n## Before the session: the only beat a human owns\n\nYou\n\nSign into chatgpt.com and gemini.google.com in your normal Chrome — then hand the keys over:\n\n``` text\noracle --engine browser -p \"HI\"\n```\n\nAgent\n\nOracle copies the session cookies from that logged-in profile into a temporary automation profile, opens ChatGPT in an automated Chrome, sends the prompt, and records the reply as a session. Any reply proves the cookie path works.\n\nCookies are the entire auth story. No API keys — the browser authenticates as *you* and gets everything the subscription gives you: the model picker, extended thinking, and the upload widget the video pipeline depends on.\n\n## The session, beat by beat\n\nYou\n\n`$oracle-video https://www.youtube.com/watch?v=uC6slhCTXlA` — Built this up into an extensive `$here-now` workshop wiki with extensive resources/pull quotes/and ai prompts to copy/paste\n\nThat is the whole human contribution to the run — one prompt naming a public 93:49 YouTube workshop (\"Building Durable Autonomous Agents\", the Loopcraft project). Everything below is the agent working.\n\nAgent\n\nSized up the source: 93 minutes and 49 seconds of screen-heavy workshop footage. No text model can watch that — only Gemini genuinely watches video, and only through the page's own upload widget. Plan: cut the recording into source-native shards Gemini can watch one at a time.\n\nThe first decision is a routing decision: text questions go to GPT-5.6 Pro, footage goes to Gemini — both through the same browser hands.\n\nAgent\n\nBefore touching video, asked the strongest text model to review the plan. Session `loopcraft-segmentation-review` (gpt-5.6-pro, completed): GPT-5.6 Pro reviewed how to shard a 93:49 recording so Gemini gets watchable, source-native chunks.\n\nPlanning the pipeline with the pipeline: maximum intelligence supervises the video intelligence before a single expensive run starts.\n\nAgent\n\nProbed before spending. A battery of cheap sentinel runs on gemini-3-pro, each asking for exactly one line back:\n\n``` text\nloopcraft-video-smoke-probe      error\nloopcraft-video-probe-retry      error\nloopcraft-video-probe-rebuilt    completed\nloopcraft-submit-proof-probe     completed\nloopcraft-awaited-upload-probe   completed\n  └─ reply: ATTACHED_VIDEO_OK loopcraft-awaited-upload-probe\n```\n\nTwo failures cost seconds each. Only after the sentinel came back byte-exact — from the real MP4 — did the long runs start. The generalized probe, liftable:\n\n``` text\n[video-upload-smoke-probe]\n\nThis is an attached-video readiness probe for a longer video analysis run.\n\nConfirm that you can inspect the attached video. Return exactly three short bullets:\n- whether video frames are visible;\n- whether audio is detectable;\n- one strictly observed visual detail, without guessing.\n\nEnd with this exact sentinel on its own line:\nEND_OF_VIDEO_ANALYSIS video-upload-smoke-probe probe\n```\n\n``` text\noracle --engine browser --model gemini-3-pro --browser-attachments always \\\n  -p \"$(cat probe-prompt.md)\" --file \"/absolute/path/to/probe-clip.mp4\" \\\n  --slug \"video-upload-smoke-probe\"\n```\n\nProbes before spends: a falsifiable one-line sentinel proves the fragile path (upload, submit, encoding) before an hour-long analysis bets on it. A paraphrase or silence means the path is broken.\n\nAgent\n\nWatched the shards. Sessions `loopcraft-workshop-shard-001…004` (gemini-3-pro): each shard analyzed for the wiki, each analysis required to end in its exact sentinel. The results prove Gemini genuinely watched the footage — it heard \"we are going to talk about Loopcraft…\" and identified the GitHub workshop page with the presenter's picture-in-picture at the 00:30 mark.\n\nTimestamped screen evidence — slides, terminal output, picture-in-picture — is the thing a transcript summarizer cannot produce.\n\nAgent\n\nShard 004 errored — twice — while its three siblings sailed through. That pattern rules out \"the pipeline is down\" and points at the input. Diagnosis: a media-encoding problem. An exact re-encode fixed it completely: the analysis finished in 73 seconds with the exact sentinel, and the re-encode settings were kept as the durable media recipe for future runs.\n\nFailure handled like engineering, not vibes: classify the error, change one variable, re-run the cheap proof, keep the fix.\n\nAgent\n\nVerified, then published. Pull quotes were source-checked against the analyses, the site build passed its checks with zero errors, and after publishing the agent curled the *live* URLs rather than trusting the publish receipt. The result is public and permanent: the [Loop Craft workshop wiki](https://silken-ferry-63mw.here.now/) and its [video companion](https://silken-ferry-63mw.here.now/video-companion/).\n\nOne working session: a 93-minute recording became a quotable artifact anyone in the room can open right now — not a chat log.\n\n## Where it landed: the shard prompt, verbatim\n\nThis exact prompt produced shard 001 of the wiki — it is the money artifact of the run. Swap the metadata block and the sentinel for your own video; keep the evidence rules, because they are what stop Gemini from inventing commands, URLs, and speakers.\n\n``` text\n[loopcraft-workshop-shard-001-source-native]\n\nAnalyze the attached source-native video shard for an extensive public workshop wiki.\n\nMetadata:\n- run_key: loopcraft-agent-workshop\n- shard_id: 001\n- original recording range: 00:00:00–00:04:00\n- publication chapter: Workshop Setup and the Looping Resolver\n\nEvidence rules:\n1. Use original workshop timestamps from 00:00:00 through 00:04:00.\n2. Separate spoken evidence, visible screen evidence, and inference.\n3. Listen to the audio; do not rely on automatic captions.\n4. Do not invent commands, URLs, files, tools, speakers, or UI labels.\n5. Pull quotes must be brief (under 20 words), audio-verified, attributed to Joel Hooks, and timestamped.\n6. Exclude credentials, account details, local paths/usernames, private URLs, signed URLs, notifications, or incidental personal data.\n7. Treat opinions and live-demo observations as such, not universal facts.\n8. Derived prompts must use placeholders, be clearly labeled as inspired by the workshop, and require confirmation for destructive or externally visible actions.\n\nReturn dense Markdown with these sections:\n# Shard 001 — 00:00:00–00:04:00\n## Executive Takeaway\n## Dense Timeline\n## Concepts and Mental Models\n## Visible Demo Evidence\n## Architecture and Workflow Patterns\n## Tools, Commands, and References\n## Pull Quote Candidates\n## Copy/Paste AI Prompt Candidates\n## Student Exercises and Reflection Questions\n## Keyframes to Capture\n## Uncertainties and Publication Review\n## Machine JSON\n\nThe fenced JSON object must contain: shard_id, original_range, summary, timeline, concepts, workflows, demos, resources, quotes, prompts, exercises, keyframes, warnings, unanswered_questions.\n\nEnd with this exact sentinel on its own line:\nEND_OF_VIDEO_ANALYSIS loopcraft-agent-workshop shard-001\n```\n\nFor the text-consult side of this section — the maximum-batch consult prompt and the full agent setup — go back to the [section guide](https://shiny-rosette-m6zy.here.now/index.html), or hear the skeptical version in the [dialogue](https://shiny-rosette-m6zy.here.now/dialogue.html). The full Oracle skill family behind this run is [attached on the index](https://shiny-rosette-m6zy.here.now/index.html#attachments-the-oracle-skill-family), downloadable and installable."
}
