Index

Mobius AI Encyclopedia¶

Status: Alpha • Backed by ECHO + Civic Ledger • GI-gated

The Mobius AI Encyclopedia is a high-integrity, auto-growing knowledge base that stores GI-scored, multi-engine consensus answers as canonical entries.

It is not a generic wiki.

It is a Civic Encyclopedia:

Every entry is:
Proposed by AI (multi-engine Thought Broker)
Evaluated by Sentinels (Claude / GPT / Gemini / DeepSeek)
Scored by GI (Global Integrity)
Optionally attested into the Civic Ledger
Only entries above a GI threshold (e.g. ≥ 0.95) become CANONICAL.
Everything else enters as DRAFT or NEEDS_REVIEW.

This is how Mobius turns raw AI output into stable, reusable, low-hallucination knowledge.

Core Concepts¶

Topic ID¶

Each entry is anchored by a stable topicId, e.g.:

civic_os.mic_tokenomics
civic_os.echo_layer
mobius.thought_broker
hive.lore.crown_bearers
usecase.boulder.climate_policy

topicId is:

Machine-friendly
Stable across time
The lookup key for all agents

Status¶

Each entry has a status:

CANONICAL – high-GI, consensus answer, safe to reuse as truth
DRAFT – decent answer but below GI threshold, needs human review
NEEDS_REVIEW – flagged for re-examination (human or Sentinel-initiated)

GI Thresholds¶

Default thresholds (configurable):

GI ≥ 0.95 → auto-eligible for CANONICAL
0.90 ≤ GI < 0.95 → DRAFT, must be reviewed
GI < 0.90 → not ingested (only logged for analysis)

Data Model (Conceptual)¶

Table: encyclopedia_entries

id: UUID
topic_id: TEXT
title: TEXT
summary: TEXT
content: TEXT (markdown encouraged)
gi_score: NUMERIC(4,3)
status: 'CANONICAL' | 'DRAFT' | 'NEEDS_REVIEW'
engines: JSONB
[{ name, version, vote, confidence }]
sources: JSONB
[{ kind, url?, docId?, title?, citation?, hash? }]
version: INTEGER
ledger_tx_id: TEXT | NULL
created_by: TEXT ('system:nightly-cron', user id, etc.)
created_at: timestamptz
updated_at: timestamptz
last_reviewed_at: timestamptz | NULL

Invariants:

(topic_id, version) is unique
Highest version with status='CANONICAL' is the current canonical entry

API Overview¶

Base path: /v1/encyclopedia

GET /v1/encyclopedia/:topicId¶

Fetch the current canonical entry for a topic.

Request:

GET /v1/encyclopedia/civic_os.mic_tokenomics

Response (200):

{
  "entry": {
    "id": "uuid",
    "topicId": "civic_os.mic_tokenomics",
    "title": "MIC Tokenomics for Citizens",
    "summary": "MIC rewards nodes for high Global Integrity scores...",
    "content": "## MIC Tokenomics\n\nLong markdown...",
    "giScore": 0.972,
    "status": "CANONICAL",
    "engines": [
      { "name": "claude", "version": "opus-4.5", "vote": "APPROVE", "confidence": 0.96 },
      { "name": "gpt", "version": "5.1", "vote": "APPROVE", "confidence": 0.98 }
    ],
    "sources": [
      {
        "kind": "doc",
        "title": "Mobius Systems: MIC Tokenomics",
        "citation": "Mobius Systems Whitepaper (2025)",
        "hash": "sha256:..."
      }
    ],
    "version": 3,
    "ledgerTxId": "ledger_tx_abc123",
    "createdBy": "system:nightly-cron",
    "createdAt": "2025-11-25T02:05:00Z",
    "updatedAt": "2025-11-25T02:05:00Z",
    "lastReviewedAt": "2025-11-25T12:00:00Z"
  }
}

Response (404):

{
  "error": "ENTRY_NOT_FOUND",
  "topicId": "civic_os.unknown"
}

GET /v1/encyclopedia?query=…

Search entries by topic/title/summary.

Request:

GET /v1/encyclopedia?query=tokenomics

Response:

{
  "results": [
    { "id": "...", "topicId": "civic_os.mic_tokenomics", "title": "...", "giScore": 0.972, "status": "CANONICAL", ... },
    { "id": "...", "topicId": "civic_os.mic_overview", "title": "...", "giScore": 0.941, "status": "DRAFT", ... }
  ]
}

POST /v1/encyclopedia/ingest¶

Internal endpoint, intended for:

Nightly cron jobs
n8n flows
System-level ingestion

Expected payload:

POST /v1/encyclopedia/ingest
{
  "topicId": "civic_os.mic_tokenomics",
  "title": "MIC Tokenomics for Citizens (Grade 10)",
  "summary": "MIC is a Global Integrity–linked reward token...",
  "content": "# MIC Tokenomics\n\nLong form markdown answer...",
  "giScore": 0.972,
  "engines": [
    { "name": "claude", "version": "opus-4.5", "vote": "APPROVE", "confidence": 0.96 },
    { "name": "gpt", "version": "5.1", "vote": "APPROVE", "confidence": 0.98 },
    { "name": "deepseek", "version": "coder", "vote": "APPROVE", "confidence": 0.95 }
  ],
  "sources": [
    {
      "kind": "doc",
      "title": "Mobius Systems: MIC Tokenomics",
      "citation": "Mobius Systems Whitepaper (2025)",
      "hash": "sha256:..."
    },
    {
      "kind": "url",
      "url": "https://example.com/mic-design",
      "title": "MIC Design Notes",
      "hash": "sha256:..."
    }
  ],
  "ledgerTxId": "ledger_tx_abc123",
  "createdBy": "system:nightly-cron"
}

Response:

{
  "entry": {
    "id": "uuid",
    "topicId": "civic_os.mic_tokenomics",
    "version": 3,
    "status": "CANONICAL",
    "giScore": 0.972,
    ...
  }
}

Status logic:

If giScore ≥ CANONICAL_THRESHOLD → stored as CANONICAL
Else → stored as DRAFT (or NEEDS_REVIEW, depending on ingest policy)

Nightly Encyclopedia Build (n8n / Cron Flow)¶

A typical nightly job:

Cron (02:00) – trigger
Topic generator – generate ~20 topics/questions:
Either static list
Or small-model “topic proposer” using repo/docs context
Thought Broker call (/v1/deliberate):
Mode: "encyclopedia_build"
Sentinels: Claude, GPT, Gemini, DeepSeek
Requires: GI score, engine votes, sources
GI filter:
If giScore ≥ 0.95 → candidate for CANONICAL
If 0.90 ≤ giScore < 0.95 → DRAFT only
Civic Ledger attestation (optional):
POST /v1/ledger/attest with content hash, GI, sources
Receive ledgerTxId
Encyclopedia ingest:
POST /v1/encyclopedia/ingest with consensus answer
Human review (optional):
If status = DRAFT or GI below threshold → send to human queue (Telegram/Discord)

Over time, this creates a growing, integrity-filtered canon of entries.

How Agents Should Use the Encyclopedia¶

Recommended retrieval priority:

ECHO Layer (short-term context & cached Q/A)
Mobius Encyclopedia (/v1/encyclopedia/:topicId)
Fresh multi-engine reasoning via Thought Broker (/v1/deliberate)

Pseudocode:

async function answerQuestion(q: QuestionInput): Promise<Answer> {
  const topicId = deriveTopicIdFromQuestion(q);

  // 1) Try ECHO (fast, short-term cache)
  const echoHit = await echo.getBestMatch(topicId, q);
  if (echoHit && echoHit.giScore >= 0.93) {
    return {
      mode: 'echo',
      source: 'ECHO',
      content: echoHit.answer,
      giScore: echoHit.giScore,
    };
  }

  // 2) Try Encyclopedia (stable, curated canon)
  const canonical = await encyclopedia.getCanonical(topicId);
  if (canonical && canonical.giScore >= 0.95) {
    return {
      mode: 'encyclopedia',
      source: 'ENCYCLOPEDIA',
      content: canonical.content,
      giScore: canonical.giScore,
      topicId,
      version: canonical.version,
    };
  }

  // 3) Fall back to fresh multi-engine reasoning
  const deliberation = await thoughtBroker.deliberate({
    prompt: q.text,
    topicId,
    requireSources: true,
    requireGiScore: true,
  });

  // Optional: if high GI, enqueue for ingest
  if (deliberation.giScore >= 0.95) {
    await encyclopedia.ingestFromDeliberation(topicId, deliberation);
  }

  return {
    mode: 'fresh',
    source: 'THOUGHT_BROKER',
    content: deliberation.consensus.answer,
    giScore: deliberation.giScore,
  };
}

Design Goals¶

Reduce hallucinations by reusing high-GI answers instead of re-inventing them every time.
Preserve provenance via engines + sources + ledger tx id.
Stay adaptive: entries can be superseded by newer, higher-GI versions.
Remain civic: humans can review, flag, or amend entries via separate workflows.

For civic-level rules around what qualifies as CANONICAL, see:

docs/encyclopedia/CHARTER.md