{
  "schemaVersion": "1",
  "revision": "22ce5d34dbcb5a78",
  "id": "codex-model-council",
  "slug": "model-council",
  "title": "Use a model council for hard decisions",
  "summary": "A single model can give a clean answer and still miss the important edge. Use a model council when the cost of a missed perspective is higher than the cost of waiting.",
  "category": "Verification",
  "tags": [
    "fusion",
    "review",
    "disagreement"
  ],
  "sources": [
    {
      "label": "Codex workflows companion",
      "url": "https://lapis-fossil-vsbc.here.now/#model-council"
    },
    {
      "label": "Published workshop recording",
      "url": "https://www.dropbox.com/scl/fi/3m5p2dwkx0imuyu0dq5os/GMT20260626-155513_Recording_gallery_1920x1080.mp4?rlkey=60j5649let6pt4fdxpggozv1q&st=qhbuo513&dl=0"
    }
  ],
  "sourceUnits": [
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/Use It For Hard Decisions",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/The Mental Model",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/Run The Whole Council",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/Handle Missing Voices",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/Good Software Uses",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/Optional Pattern: Stop-Hook Review",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/Copyable Artifact: Fusion Decision Brief",
    "src/content/final/chunk-09-use-a-model-council-when-one-answer-is-too-narrow.md#Use A Model Council When One Answer Is Too Narrow/Claim Boundaries",
    "src/lib/content/sections.ts",
    "src/lib/content/sections.ts#model-council",
    "src/lib/content/sections.ts#recordingSpans"
  ],
  "collection": "codex",
  "categoryId": "verification",
  "wordCount": 867,
  "url": "/topics/codex-model-council/",
  "markdownUrl": "/markdown/codex-model-council.md?raw=1&v=22ce5d34dbcb5a78",
  "jsonUrl": "/api/v1/documents/codex-model-council.json",
  "headings": [
    {
      "id": "use-it-for-hard-decisions",
      "text": "Use It For Hard Decisions",
      "level": 2
    },
    {
      "id": "the-mental-model",
      "text": "The Mental Model",
      "level": 2
    },
    {
      "id": "run-the-whole-council",
      "text": "Run The Whole Council",
      "level": 2
    },
    {
      "id": "handle-missing-voices",
      "text": "Handle Missing Voices",
      "level": 2
    },
    {
      "id": "good-software-uses",
      "text": "Good Software Uses",
      "level": 2
    },
    {
      "id": "optional-pattern-stop-hook-review",
      "text": "Optional Pattern: Stop-Hook Review",
      "level": 2
    },
    {
      "id": "copyable-artifact-fusion-decision-brief",
      "text": "Copyable Artifact: Fusion Decision Brief",
      "level": 2
    },
    {
      "id": "claim-boundaries",
      "text": "Claim Boundaries",
      "level": 2
    },
    {
      "id": "related-resources",
      "text": "Related resources",
      "level": 2
    }
  ],
  "markdown": "# Use a model council for hard decisions\n\nA single model can give a clean answer and still miss the important edge. Use a **model council** when the cost of a missed perspective is higher than the cost of waiting.\n\n**[Fusion](https://github.com/johnlindquist/fusion)** is one way to run that pattern locally: send the same bounded software question to multiple configured providers, wait for the panel to finish, let a **finalizer** compare the outputs, then make the engineering decision from the result and the repo evidence.\n\n> It can take all of those results and then fuse them with the model of your choice. Meaning that… you get multiple opinions, you can have them kind of argue against each other.\n>\n> — John Lindquist\n\n## Use It For Hard Decisions\n\nReach for a council when you need **coverage, not speed**.\n\nGood fits:\n\n- large PR review\n- architecture tradeoffs\n- migration risk\n- security and performance tension\n- uncertain debugging plans\n- product or UX direction\n- release or rollback strategy\n\nDo not use it as a way to avoid making the call. The council gives you independent critique. **You still own the decision.**\n\n## The Mental Model\n\n```text\nbounded software question\n-> independent model/provider perspectives\n-> terminal provider states\n-> finalizer comparison\n-> summary.json\n-> final.md\n-> engineering decision\n```\n\nFusion is for coverage, not voting. The useful output is not \"three models said yes.\" The useful output is consensus, contradictions, unique warnings, blind spots, and a final recommendation you can test against the codebase.\n\n## Run The Whole Council\n\nPartial panel output is not the answer. Wait until every configured provider reaches a **terminal state** before treating the result as complete.\n\nBefore a real run:\n\n- inspect the configured providers\n- run the full configured panel by default\n- wait for terminal provider states\n- read `summary.json`\n- read `final.md`\n- reconcile the result with repo evidence\n\n**Panel artifacts are partial until the final artifacts exist.**\n\n## Handle Missing Voices\n\nA missing provider degrades the council. It does not automatically make the run useless.\n\nIf a provider times out, fails, or hits quota after artifacts exist, treat that as a **missing voice**. The final answer can still be useful if it clearly names the degraded state and enough other panel output exists to synthesize.\n\nDo not hide the missing voice. Make it part of the decision record.\n\n## Good Software Uses\n\nLarge PR review:\n\n```text\nUse Fusion to review this PR before merge.\n\nContext:\n- PR summary:\n- changed files:\n- risky areas:\n- test output:\n- migration or rollout notes:\n\nAsk the council to find:\n- correctness risks\n- security risks\n- performance risks\n- missing tests\n- rollout or rollback issues\n- documentation gaps\n\nReturn:\n- consensus\n- contradictions\n- unique warnings\n- blind spots\n- recommended merge/block decision\n- verification commands\n```\n\nArchitecture tradeoff:\n\n```text\nCompare three implementation paths for the billing webhook fix:\n\nA. provider event id dedupe\nB. invoice-level lock\nC. processing ledger\n\nInclude:\n- failure modes\n- migration risk\n- rollback path\n- test plan\n- operational logging\n- smallest shippable recommendation\n```\n\nRole/perspective panel:\n\n```text\nReview this auth/session plan from four perspectives:\n\n1. security reviewer\n2. performance reviewer\n3. migration reviewer\n4. product/UX reviewer\n\nEach reviewer should name what could go wrong.\nThe finalizer should preserve minority warnings instead of flattening them.\n```\n\n## Optional Pattern: Stop-Hook Review\n\nFor risky AFK work, a prompt marker can ask a later stop gate to review changes before reporting done.\n\nConceptual shape:\n\n```text\nWhen the user prompt includes #review, write a marker for the stop hook.\n\nAt Stop:\n- check whether files changed\n- check whether the marker exists\n- run Fusion against the diff\n- read summary.json and final.md\n- report the council result before claiming done\n```\n\nTreat this as a pattern until the hook payload, marker passing, and current CLI behavior are verified.\n\n## Copyable Artifact: Fusion Decision Brief\n\n```text\nDecision:\n[the software decision under review]\n\nWhy a council:\n[what one model might miss]\n\nContext:\n- repo area:\n- changed files:\n- relevant tests:\n- logs or traces:\n- constraints:\n\nProvider instructions:\nAsk each provider to review independently.\nPreserve disagreements.\nDo not treat majority agreement as proof.\n\nFinalizer instructions:\nReturn:\n- consensus\n- contradictions\n- unique warnings\n- blind spots\n- recommended decision\n- verification commands\n- confidence and why\n\nHuman decision:\n[accept / revise / block / run another pass]\n```\n\n## Claim Boundaries\n\nHosted products may publish their own benchmark or cost claims for panel-style research workflows. Treat those as vendor claims that need dated citations. Local Fusion is a local CLI pattern inspired by [the same panel/finalizer idea](https://openrouter.ai/blog/announcements/fusion-beats-frontier/), not proof of hosted benchmark results.\n\n> I can, like, anecdotally vouch for the results, but I can't give you some sort of percentage of… it's this close to Fable or whatever.\n>\n> — John Lindquist\n\nProvider availability, pricing, quotas, auth, and data handling are local/account-dependent. Verify current provider terms before sending private code or prompts.\n\nModel council output is **advisory**. Tests, code review, product judgment, and repo evidence still decide the work.\n\n## Related resources\n\n- [Companion reference](https://lapis-fossil-vsbc.here.now/#model-council)\n- [Resource directory and portable download](/topics/codex-resources/)\n- [Published recording](https://www.dropbox.com/scl/fi/3m5p2dwkx0imuyu0dq5os/GMT20260626-155513_Recording_gallery_1920x1080.mp4?rlkey=60j5649let6pt4fdxpggozv1q&st=qhbuo513&dl=0)"
}
