{
  "title": "Decisions over stored embeddings",
  "revision": "Two-minute-thirty landscape cut; slower explanations preserved",
  "opening_example": {
    "state": "Please refund this charge",
    "question": "What does the customer want?",
    "options": [
      "Refund",
      "Replacement",
      "Information"
    ],
    "status": "Authored schematic; the illustrative selection is not a recorded Jev or Venn prediction.",
    "jev_reference": "https://docs.typesafe.ai/primitives/choice.md",
    "architecture_reference": "https://docs.typesafe.ai/concepts/system-one.md"
  },
  "status": "Local review draft; clearances granted by owner; do not publish without explicit instruction.",
  "narration": false,
  "music": "Original synth score in A minor (Am-F-C-G, 118 BPM): pads, arpeggio, bass, soft kick/clap groove and a pentatonic lead that build with the sections; rising swells at cuts and in-key bell chimes for retrieval; no samples or third-party recordings.",
  "speech_playback": false,
  "source_narratives_or_wine_notes": false,
  "on_screen_fine_print": false,
  "minimum_text_size_at_1080_output": 48,
  "layout": "Padded answer-box containment, metric-column bounds, reserved narration area and per-frame text-overlap checks",
  "narration_style": "One cool-blue regular-weight caption block, distinct from graphic labels, without sentence-ending periods",
  "disclosure": "On methods page only: Microsoft employee; personal research. Owner confirmed clearances on 2026-11-07; publication on hold.",
  "animation": "Continuous 30 fps record conveyors, camera-depth motion, data morphs, retrieval trails, archive wipe and processing lanes. Slower architecture and filing-label reveals; no frozen post-intro frames.",
  "timing": "All animation speeds are editorial, not real-time execution or measurement recordings.",
  "pacing": "Landscape trimmed to 150 seconds by removing the 3-second summary animation and shortening the end card by 1 second; vertical remains 60 seconds. Slower architecture stages and the 10-second code highlight are unchanged.",
  "audio_section": false,
  "evidence": {
    "history-eval.json": "c32631689cf74c6ecc92e28f773ee83c05aadbe8c20208ef207170e441a1fdc1",
    "sec-eval.json": "37034936e0ed6142eca99cf07b84f424605ccc67e334a6a56397183f788c211f",
    "wine-eval.json": "583ed6e6dcd1b8763f912c0e86481fbc336e692822c846d3b673d376c33c1121"
  },
  "head_execution": {
    "file": "same-memory/head-execution.json",
    "sha256": "81b0770a2e6fb59cabe5efeb58fba238bfc24b0d62cd8def61f052990f35d47a",
    "source": "examples/same_memory_video/head_demo.py",
    "source_sha256": "e253fe7079f2843e908969276b8ce73f1997eb5eb22415e3aa93a25730cea9d7",
    "replay": "Verbatim executed source; replay compressed to fit the edit, without a real-time clock."
  },
  "renderer_sha256": "c4f3db85f3468b850f65a0ac41b5dd835fc576451c36b611545df6fdd7d919ff",
  "media": [
    {
      "file": "venn-same-memory-review.mp4",
      "sha256": "54ad1ca21b79323b8f78ef40d8f421baccf2536352cf108ec0b64dc06fe1ccc5",
      "bytes": 16547056,
      "seconds": 150,
      "width": 1920,
      "height": 1080,
      "fps": 30
    },
    {
      "file": "venn-same-memory-vertical-review.mp4",
      "sha256": "3d3e8165be02dca8d8fed7796de1f52dc7546300611afc7cf933b0a8d374aeaa",
      "bytes": 4395562,
      "seconds": 60,
      "width": 1080,
      "height": 1920,
      "fps": 30
    }
  ],
  "cuts": {
    "landscape": [
      {
        "start": 0,
        "end": 4,
        "kind": "setup",
        "card": "Both choose from answers supplied with your question"
      },
      {
        "start": 4,
        "end": 13,
        "kind": "selection",
        "card": "Jev returns a selected answer and its probabilities",
        "model": "Jev"
      },
      {
        "start": 13,
        "end": 24,
        "kind": "selection",
        "card": "Venn scores answers using numeric representations of meaning",
        "model": "Venn"
      },
      {
        "start": 24,
        "end": 28,
        "kind": "history",
        "card": "Now predict outcomes from complaints to this company",
        "mode": "question"
      },
      {
        "start": 28,
        "end": 34,
        "kind": "history",
        "card": "Use Jev's label to group past outcomes",
        "mode": "baseline"
      },
      {
        "start": 34,
        "end": 42,
        "kind": "history",
        "card": "Venn retrieves similar complaints with recorded outcomes",
        "mode": "retrieve"
      },
      {
        "start": 42,
        "end": 48,
        "kind": "history",
        "card": "Evaluate outcome rankings on previously unseen complaints",
        "mode": "primary"
      },
      {
        "start": 48,
        "end": 52,
        "kind": "history",
        "card": "Now compare only complaints requesting a refund",
        "mode": "refund"
      },
      {
        "start": 52,
        "end": 57,
        "kind": "history",
        "card": "Imagine removing the complaint archive",
        "mode": "wipe"
      },
      {
        "start": 57,
        "end": 64,
        "kind": "sec",
        "card": "Now learn industry labels from company filings",
        "mode": "learn"
      },
      {
        "start": 64,
        "end": 69,
        "kind": "sec",
        "card": "Test both methods on the cookbook's filings",
        "mode": "sample"
      },
      {
        "start": 69,
        "end": 75,
        "kind": "sec",
        "card": "Then test on additional held-out company filings",
        "mode": "confirmation"
      },
      {
        "start": 75,
        "end": 81,
        "kind": "sec",
        "card": "Inspect one filing's predictions",
        "mode": "example"
      },
      {
        "start": 81,
        "end": 89,
        "kind": "sec",
        "card": "Venn learns industry labels from past filings",
        "mode": "neighbors"
      },
      {
        "start": 89,
        "end": 94,
        "kind": "work",
        "card": "Predict the critic's score from wine tasting notes",
        "mode": "question"
      },
      {
        "start": 94,
        "end": 102,
        "kind": "work",
        "card": "Jev builds features by answering questions about notes",
        "mode": "flow"
      },
      {
        "start": 102,
        "end": 116,
        "kind": "code",
        "card": "Venn fits a prediction head on stored embeddings"
      },
      {
        "start": 116,
        "end": 122,
        "kind": "work",
        "card": "Evaluate critic scores on previously unseen wine notes",
        "mode": "accuracy"
      },
      {
        "start": 122,
        "end": 127,
        "kind": "work",
        "card": "Both learn from the same 1,200 training notes",
        "mode": "time"
      },
      {
        "start": 127,
        "end": 134,
        "kind": "price",
        "card": "Price uses a different, already-stored embedding cache",
        "mode": "question"
      },
      {
        "start": 134,
        "end": 140,
        "kind": "price",
        "card": "Venn's time excludes the already-completed embedding step",
        "mode": "time"
      },
      {
        "start": 140,
        "end": 143,
        "kind": "price",
        "card": "Local predictions require no additional API charges",
        "mode": "cost"
      },
      {
        "start": 143,
        "end": 147,
        "kind": "scoreboard",
        "card": "Additional tests favor Jev on these reading tasks"
      },
      {
        "start": 147,
        "end": 150,
        "kind": "end",
        "card": "Build decisions from your own data"
      }
    ],
    "vertical": [
      {
        "start": 0,
        "end": 4,
        "kind": "setup",
        "card": "Both choose from answers supplied with your question"
      },
      {
        "start": 4,
        "end": 13,
        "kind": "selection",
        "card": "Jev returns a selected answer and its probabilities",
        "model": "Jev"
      },
      {
        "start": 13,
        "end": 25,
        "kind": "selection",
        "card": "Venn scores answers using numeric representations of meaning",
        "model": "Venn"
      },
      {
        "start": 25,
        "end": 29,
        "kind": "history",
        "card": "Now predict outcomes from complaints to this company",
        "mode": "question"
      },
      {
        "start": 29,
        "end": 36,
        "kind": "history",
        "card": "Use Jev's label to group past outcomes",
        "mode": "baseline"
      },
      {
        "start": 36,
        "end": 44,
        "kind": "history",
        "card": "Venn retrieves similar complaints with recorded outcomes",
        "mode": "retrieve"
      },
      {
        "start": 44,
        "end": 50,
        "kind": "history",
        "card": "Evaluate outcome rankings on previously unseen complaints",
        "mode": "primary"
      },
      {
        "start": 50,
        "end": 54,
        "kind": "history",
        "card": "Imagine removing the complaint archive",
        "mode": "wipe"
      },
      {
        "start": 54,
        "end": 57,
        "kind": "history",
        "card": "Restore those cases to make retrieval useful again",
        "mode": "restore"
      },
      {
        "start": 57,
        "end": 60,
        "kind": "end",
        "card": "Build decisions from your own data"
      }
    ]
  },
  "limitations": [
    "Owner confirmed clearances; public upload and publication remain on hold.",
    "SEC primary gates failed; confirmation is report-only, not a win.",
    "Wipe is conceptual; no measured empty-archive 0.50 result is claimed.",
    "Separate historical build timings; workflow motion is illustrative. Code is an executed, compressed replay, not live footage.",
    "Price uses cached Classification vectors, not the winning score head's SentenceSimilarity vectors.",
    "Notebooks verify exported point estimates, not end-to-end experiments or historical timings.",
    "Fine print and disclosures remain on the methods page at the user's request."
  ],
  "deliverables": {
    "venn-same-memory-review-captions.srt": {
      "sha256": "8895ddea35ceeae346d714bd4e50fdb480fb7965d108cac5f724c030ecf0d53f",
      "bytes": 1991
    },
    "venn-same-memory-review-captions.vtt": {
      "sha256": "af1a705f30afd7682846fd113df3f477e4c918286492a99d0a301ee7100d69d1",
      "bytes": 1999
    },
    "venn-same-memory-review-poster.jpg": {
      "sha256": "d3322af263abb6f2526fc17131a3c6267df864e88ba4622ae351731cc8beee76",
      "bytes": 108004
    },
    "venn-same-memory-review-storyboard.png": {
      "sha256": "8a2649e1b4778dd7814da3c6b4e3a5b83ce15d3669eab8cf7fc4fe127f084f7b",
      "bytes": 763180
    },
    "venn-same-memory-review.mp4": {
      "sha256": "54ad1ca21b79323b8f78ef40d8f421baccf2536352cf108ec0b64dc06fe1ccc5",
      "bytes": 16547056
    },
    "venn-same-memory-vertical-review-captions.srt": {
      "sha256": "8f0781db202d9abe112f48bc6bdf720c6cb083e9397957a6b9b4996c8e97915e",
      "bytes": 828
    },
    "venn-same-memory-vertical-review-captions.vtt": {
      "sha256": "80632cc1a0665868b6107dbd0b8e2c6b3a5d69756ddb6a3869562b5679e8bfbd",
      "bytes": 836
    },
    "venn-same-memory-vertical-review-poster.jpg": {
      "sha256": "092e73a653c3a8d171e99ae0d671ab1f196dc4bb9170c1e312db93b6f019928c",
      "bytes": 95746
    },
    "venn-same-memory-vertical-review-storyboard.png": {
      "sha256": "9bd79b2cf2452d8cd86e3fa8102078e26f433bb5c3cc8741792944e3c5f9af0d",
      "bytes": 345659
    },
    "venn-same-memory-vertical-review.mp4": {
      "sha256": "3d3e8165be02dca8d8fed7796de1f52dc7546300611afc7cf933b0a8d374aeaa",
      "bytes": 4395562
    }
  }
}