How QUORUM (MiSES) Works (Story Edition)

QUORUM (MiSES) (Quorum of Unified Observations and Referenced Underlying Material) is the evidence rule that keeps answers concise and provable by selecting a MiSES (Minimal Evidence Set) per claim: the smallest non‑redundant bundle of artefacts that still supports what the answer says.

The master plan calls QUORUM (MiSES) out alongside CROWN receipts and domain diversity (§Architecture, §Trust & Integration, §Perf). This page walks through a relatable query, shows how artefacts link together, and highlights what auditors receive.

Scenario: “Why is the sky blue?”

1. A curious users asks in verified mode.
2. Engine retrieves artefacts from NASA Climate, the U.S. National Weather Service, and a science encyclopedia.
3. QUORUM (MiSES) selection keeps only the minimal set (MiSES) that proves the claims (“sunlight scatters” and “blue travels farther”) while enforcing diverse domains.
4. Counterfactual lane looks for articles that disagree (e.g., a blog claiming the sky should appear violet) and labels them as disputed.
5. Answer ships with easy-to-read citations, timestamps, and hashes; the Web UI renders Why Trust panels, QUORUM (MiSES) chips, and playful copy that a child (or their parent) can follow.

Discovery Flow

sequenceDiagram
    participant User as Curious Kid
    participant Web as WebCrux BFF
    participant Engine as EngineCrux
    participant QuorumSel as QUORUM (MiSES) Selector

    User->>Web: Ask \"Why is the sky blue?\"
    Web->>Engine: POST /v1/answers { mode: \"verified\", k: 10 }
    Engine->>Engine: Hybrid retrieval (FTS + vector + licence signals)
    Engine->>QuorumSel: Candidate artefacts (score, domain, hash)
    QuorumSel-->>Engine: Minimal covering set + contradiction notes
    Engine-->>Web: Answer + receipt + MiSES/artefact IDs
    Web-->>User: Summary, citations, counterfactual banner

sequenceDiagram
    participant User as Curious Kid
    participant Web as WebCrux BFF
    participant Engine as EngineCrux
    participant QuorumSel as QUORUM (MiSES) Selector

    User->>Web: Ask \"Why is the sky blue?\"
    Web->>Engine: POST /v1/answers { mode: \"verified\", k: 10 }
    Engine->>Engine: Hybrid retrieval (FTS + vector + licence signals)
    Engine->>QuorumSel: Candidate artefacts (score, domain, hash)
    QuorumSel-->>Engine: Minimal covering set + contradiction notes
    Engine-->>Web: Answer + receipt + MiSES/artefact IDs
    Web-->>User: Summary, citations, counterfactual banner

The selector ensures each atomic claim is justified by at least one independent artefact, while preferring domain diversity per trust policy.

Reference Tree

graph TD
    subgraph Claim Cluster
        C1["Claim A: Sunlight scatters inside Earth's air"]
        C2["Claim B: Blue light scatters more than red light"]
    end

    subgraph Domains
        D1("nasa.gov")
        D2("weather.gov")
        D3("learnscience.fun")
        D4("skycolors.blog")
    end

    subgraph Artefacts
        A1((NASA Climate - Why is the Sky Blue? 2025))
        A2((NOAA Fact Sheet - Light & Atmosphere))
        A3((Science Encyclopedia - Color Chapter))
    end

    subgraph Counterfactuals
        CF1((Opinion blog - The sky should look violet 2019))
    end

    D1 --> A1
    D2 --> A2
    D3 --> A3
    D4 -.-> CF1

    A1 --> C1
    A2 --> C1
    A2 --> C2
    A3 --> C2
    CF1 -. disputed .-> C1

    C1 --> Z1["Citation chips: NASA (CC-BY) / NOAA (Public Domain)"]
    C1 --> Z2["Why Trust card: freshness, domain diversity, licence"]
    C1 --> Z3["Counterfactual banner: blog disagrees (WARN)"]
    C1 --> Z4["Receipts: downloadable CROWN snapshot"]

graph TD
    subgraph Claim Cluster
        C1["Claim A: Sunlight scatters inside Earth's air"]
        C2["Claim B: Blue light scatters more than red light"]
    end

    subgraph Domains
        D1("nasa.gov")
        D2("weather.gov")
        D3("learnscience.fun")
        D4("skycolors.blog")
    end

    subgraph Artefacts
        A1((NASA Climate - Why is the Sky Blue? 2025))
        A2((NOAA Fact Sheet - Light & Atmosphere))
        A3((Science Encyclopedia - Color Chapter))
    end

    subgraph Counterfactuals
        CF1((Opinion blog - The sky should look violet 2019))
    end

    D1 --> A1
    D2 --> A2
    D3 --> A3
    D4 -.-> CF1

    A1 --> C1
    A2 --> C1
    A2 --> C2
    A3 --> C2
    CF1 -. disputed .-> C1

    C1 --> Z1["Citation chips: NASA (CC-BY) / NOAA (Public Domain)"]
    C1 --> Z2["Why Trust card: freshness, domain diversity, licence"]
    C1 --> Z3["Counterfactual banner: blog disagrees (WARN)"]
    C1 --> Z4["Receipts: downloadable CROWN snapshot"]

• Solid edges = artefacts selected into the MiSES by QUORUM (MiSES).
• Dashed edge = counterfactual evidence surfaced but not part of the minimal set.
• Each artefact carries: URL or doc ID, observed timestamp, BLAKE3 hash, licence, child-friendly summary, source domain, and a freshness score recorded in the provenance ledger.

Metadata Snapshot

Sample Payload

{
  "claim_cluster_id": "sky-why-blue",
  "coverage": [
    {
      "claim_id": "C1",
      "artefact_id": "A1",
      "domain": "nasa.gov",
      "score": 0.92,
      "licence": "CC-BY",
      "observed_at": "2025-10-11T09:23:00Z"
    },
    {
      "claim_id": "C2",
      "artefact_id": "A2",
      "domain": "weather.gov",
      "score": 0.88,
      "licence": "Public Domain",
      "observed_at": "2025-10-10T13:00:00Z"
    }
  ],
  "counterfactuals": [
    {
      "claim_id": "C1",
      "artefact_id": "CF1",
      "reason": "Contradiction",
      "action": "Display as WARN",
      "score": 0.21
    }
  ]
}

After the Claim Cluster

1. Citation chips: Each claim renders two chips (“NASA • CC-BY”, “NOAA • Public Domain”) so even a non-technical reader can see who said what.
2. Why Trust card: Shows the same metadata table in miniature, highlighting freshness, domain diversity, and licence.
3. Counterfactual banner: Because the blog disagrees, WebCrux adds a yellow banner explaining that most experts say blue, while one source argued otherwise.
4. Receipts: Parents or teachers can tap “View receipt” to download the deterministic log showing the QUORUM (MiSES) graph, hashes, and cost envelope.

Storage & Audit Trail

Auditors replay the exact MiSES by loading the receipt, verifying hashes against the provenance ledger, and ensuring the set still covers every claim. If a contradiction later becomes authoritative, WatchCrux marks the affected answers WARN/FAIL and WebCrux banners them automatically.

Takeaways

• QUORUM (MiSES) keeps evidence focused: the answer cites only what is necessary.
• Counterfactuals stay visible: disagreements show up without bloating the main proof.
• Every link is auditable: hashes + receipts + WatchCrux findings close the loop.

Use this story when onboarding teammates or pitching the trust surface to customers they can see exactly how two artefacts (and beyond) underpin the claims they care about.