{"slug": "ions-a-distributed-reasoning-graph-built-from-evidence-backed-claims", "title": "Ions, a distributed reasoning graph built from evidence backed claims", "summary": "Ions, an open protocol for publishing and traversing evidence-backed claims called Cognitive Building Blocks (CBBs), has been launched. In a test, an 8B parameter model using the Ions network matched or exceeded a frontier model on 5 of 8 domain-specific queries at lower compute cost. The protocol externalizes knowledge into a traversable network, making reasoning chains visible and reusable.", "body_md": "**A lightweight protocol for publishing, connecting, and traversing Cognitive Building Blocks into reusable reasoning paths.**\n\nIONS is an open protocol that inverts the traditional AI architecture. Instead of compressing knowledge into model weights, IONS externalizes knowledge into a traversable network of atomic, typed, and evidence-backed claims called **Cognitive Building Blocks (CBBs)**. Any lightweight model can then traverse this network to produce answers with visible reasoning chains.\n\nGenesis Result: An 8B parameter model connected to a CBB network matched or exceeded a frontier model on 5 of 8 domain-specific queries — at a fraction of the compute cost.\n\nTraditional AI: `Model → Knowledge → Answer`\n\nIONS: `CBBs → Relationships → Traversal → Reasoning Path → Answer`\n\nThe durable asset is not the model. The durable asset is the network of CBBs, relationships, and reasoning paths. Models are replaceable interpreters. The network compounds.\n\nA **Cognitive Building Block** is a single atomic, assertable claim:\n\n```\n{\n  \"cbb_id\": \"cbb_a1b2c3d4e5f6\",\n  \"type\": \"claim\",\n  \"domain\": \"organizational_intelligence\",\n  \"content\": \"Organizations that automate without operational discovery build on incorrect assumptions.\",\n  \"confidence\": 0.85,\n  \"evidence\": [{\"source_type\": \"book\", \"source_id\": \"abr_chapter_3\"}],\n  \"assumptions\": [\"Organization has existing workflows to map\"],\n  \"scope\": [\"enterprise_ai\", \"process_automation\"],\n  \"status\": \"published\"\n}\n```\n\nCBBs are:\n\n**Atomic**— one claim only, specific enough to evaluate** Typed**— confidence, evidence, scope, and assumptions are explicit** Addressable**— every CBB has a stable ID and cryptographic hash** Traversable**— connected by typed relationships to other CBBs\n\nA reasoning path is the ordered result of traversal — stored as a reusable artifact:\n\n```\nQuery: \"Why do AI initiatives fail early?\"\n\nPath: cbb_001 → [supports] → cbb_014 → [causes] → cbb_027 → [depends_on] → cbb_039\nConfidence: 65.5%\nAnswer: \"Many enterprise AI initiatives fail early because shallow operational \n         discovery produces incomplete requirements...\"\n```\n\nPaths are inspectable, scorable, and reusable. The reasoning is visible — not a black box.\n\nGenesis is the first IONS node. It proves the protocol works with a curated corpus of CBBs across multiple domains including organizational intelligence, economics, blockchain, peak performance, AI regulation, and healthcare AI.\n\n| Query Type | Raw 8B | 8B + CBB Traversal | Claude Sonnet |\n|---|---|---|---|\n| Domain-specific | Weak | Strong |\nStrong |\n| Cross-domain reasoning | Weak | Competitive |\nStrong |\n| Average path confidence | — | 0.547 | — |\n\nThe 8B + CBB network matched or exceeded the frontier model on 5 of 8 domain queries.\n\n```\nKnowledge Sources (documents, books, research, observations)\n        ↓\nLight D2Brain Extractor (LLM-powered CBB extraction)\n        ↓\nReview Queue (human approval before publication)\n        ↓\nIONS Network\n  ├── CBB Registry\n  ├── Relationship Registry  \n  ├── Path Registry\n  └── Traversal Engine\n        ↓\nQuery Interface → Answer + Reasoning Path + Evidence\n```\n\nThe reference genesis node is publicly accessible. You can query it directly or register your node against it:\n\n```\n# Query the genesis node\ncurl -X POST http://162.243.203.243:8000/query \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\"query\": \"How does institutional memory compound competitive advantage?\"}'\n\n# Register your node with genesis\ncurl -X POST http://162.243.203.243:8000/nodes/register \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\"node_id\": \"your_node_id\", \"public_api_base\": \"https://your-node.example.com\"}'\n\n# Node manifest\ncurl http://162.243.203.243:8000/.well-known/ions-node.json\n```\n\n- Docker and Docker Compose\n- An\n[OpenRouter](https://openrouter.ai)API key (for CBB extraction and relationship generation) - Node.js 18+ (for the frontend)\n\n```\ngit clone https://github.com/ions-protocol/genesis\ncd genesis\ncp .env.example .env\n```\n\nEdit `.env`\n\n:\n\n```\nOPENROUTER_API_KEY=sk-or-your-key-here\nIONS_NODE_ID=my_node\nDATABASE_URL=postgresql+asyncpg://ions:ions@localhost:5432/ions\ndocker compose up -d\n```\n\nThis starts:\n\n- PostgreSQL (port 5432) — system of record\n- FastAPI backend (port 8000) — CBB registry, traversal engine, API\n- Next.js frontend (port 3000) — explorer, workbench, graph\n\n```\n# Health check\ncurl http://localhost:8000/health\n\n# Node manifest\ncurl http://localhost:8000/.well-known/ions-node.json\n\n# API docs\nopen http://localhost:8000/docs\n\n# Frontend\nopen http://localhost:3000\n```\n\n- Open\n`http://localhost:3000/contribute`\n\n- Paste text or upload a\n`.txt`\n\n,`.md`\n\n, or`.docx`\n\ndocument - Click\n**Extract CBBs**— the network identifies atomic claims automatically - Review and deselect any you don't want\n- Submit to the review queue\n- Approve in the Workbench\n\n```\ncurl -X POST http://localhost:8000/cbb \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\n    \"type\": \"claim\",\n    \"domain\": \"organizational_intelligence\",\n    \"content\": \"Shallow discovery is the most common cause of failed AI transformation.\",\n    \"confidence\": 0.85,\n    \"evidence\": [{\"source_type\": \"original\", \"source_id\": \"field_observation_2024\"}],\n    \"status\": \"candidate\"\n  }'\n```\n\n✓ **Good**: \"Organizations that skip operational discovery produce AI requirements that don't match how work actually happens.\"\n\n✗ **Too vague**: \"AI is important for business.\"\n\n✗ **Compound claim**: \"AI fails because of bad data and poor leadership and unclear goals.\"\n\nOne claim. Specific enough to evaluate as true or false in context.\n\n```\ncurl -X POST http://localhost:8000/query \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\n    \"query\": \"Why do institutional knowledge gaps compound over time?\",\n    \"top_n_paths\": 3,\n    \"max_depth\": 5\n  }'\n```\n\nResponse includes:\n\n`raw_answer`\n\n— direct LLM answer without CBB context`cbb_answer`\n\n— answer synthesized from traversal paths`paths`\n\n— ordered reasoning paths with confidence scores, CBB sequences, and relationship chains\n\n| Method | Endpoint | Description |\n|---|---|---|\n| POST | `/cbb` |\nPublish a CBB |\n| GET | `/cbb` |\nSearch CBBs |\n| GET | `/cbb/{id}` |\nRetrieve a CBB |\n| POST | `/cbb/{id}/deprecate` |\nDeprecate a CBB |\n| POST | `/relationship` |\nCreate a relationship |\n| GET | `/relationship` |\nList relationships |\n| POST | `/query` |\nRun traversal query |\n| GET | `/path/{id}` |\nRetrieve a saved path |\n| GET | `/path` |\nList saved paths |\n| GET | `/health` |\nNode health check |\n| GET | `/.well-known/ions-node.json` |\nNode manifest |\n| GET | `/stats` |\nNetwork statistics |\n| GET | `/docs` |\nInteractive API docs |\n\nEvery IONS node is independent. To run a node:\n\n- Fork this repository\n- Configure your\n`.env`\n\n- Run\n`docker compose up -d`\n\n- Seed your node with CBBs from your domain\n- Your node announces itself via\n`/.well-known/ions-node.json`\n\nYour node's manifest is automatically available at:\n\n```\nGET /.well-known/ions-node.json\n{\n  \"node_id\": \"your_node_id\",\n  \"protocol_version\": \"ions-genesis-0.1\",\n  \"supported_cbb_types\": [\"claim\"],\n  \"capabilities\": [\"publish_cbb\", \"publish_relationship\", \"query\", \"traverse\"],\n  \"public_api_base\": \"https://your-node-url.com\",\n  \"status\": \"active\",\n  \"open_contributions\": true\n}\n```\n\nUpdate `public_api_base`\n\nin `backend/main.py`\n\nto your public URL before deploying.\n\nOnce your node is running at a public URL, register it with any existing IONS node:\n\n```\ncurl -X POST https://known-node.example.com/nodes/register \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\"node_id\": \"your_node_id\", \"public_api_base\": \"https://your-node.example.com\"}'\n```\n\nYour node will be discovered via its manifest, added to the registry, and queries on that node will automatically traverse your CBBs as part of federated paths.\n\nAfter publishing CBBs, generate relationships between them:\n\n```\nsource .env\npython3 generate_relationships_fast.py\n```\n\nThis connects your CBBs to each other using the same LLM-powered approach used to build the Genesis network. Run multiple times to increase relationship density. Aim for 3+ relationships per CBB for strong traversal paths.\n\nNarrow Super Intelligence clusters are domain-specific groupings of CBBs. The graph view automatically groups your CBBs into NSIs using LLM-powered semantic clustering. As nodes contribute knowledge in new domains, new NSI clusters form organically.\n\nPath confidence is computed from:\n\n```\nPathConfidence = CBB_avg × REL_avg × EvidenceScore\n```\n\nConfidence is always shown with its path and evidence — never as a context-free number.\n\n| Type | Meaning |\n|---|---|\n`supports` |\nSource increases confidence of target |\n`contradicts` |\nSource challenges target |\n`depends_on` |\nSource requires target as premise |\n`causes` |\nSource leads to target |\n`correlates_with` |\nSource and target are associated |\n`extends` |\nSource expands target |\n`refines` |\nSource narrows or clarifies target |\n`references` |\nSource cites target |\n\n```\nions-genesis/\n├── backend/\n│   ├── main.py                 # FastAPI app, health, manifest\n│   ├── app/\n│   │   ├── api/\n│   │   │   ├── cbbs.py         # CBB CRUD endpoints\n│   │   │   ├── relationships.py # Relationship endpoints  \n│   │   │   └── query.py        # Traversal and path endpoints\n│   │   ├── models/\n│   │   │   ├── schemas.py      # Pydantic validation models\n│   │   │   └── artifacts.py    # SQLAlchemy ORM models\n│   │   ├── services/\n│   │   │   ├── traversal.py    # Path enumeration and scoring\n│   │   │   ├── synthesis.py    # Answer generation\n│   │   │   └── hashing.py      # Canonical hash service\n│   │   └── core/\n│   │       ├── config.py       # Settings\n│   │       └── database.py     # Async DB connection\n│   └── seed.py                 # Initial seed data\n├── frontend/\n│   └── app/                    # Next.js pages\n│       ├── page.tsx            # Explorer\n│       ├── graph/              # NSI graph visualization\n│       ├── contribute/         # CBB contribution\n│       ├── workbench/          # Review and approval\n│       ├── rights/             # Attribution (coming soon)\n│       ├── node/               # Node status and manifest\n│       └── settings/           # API key and model config\n├── generate_relationships_fast.py  # Bulk relationship generation\n├── docker-compose.yml\n└── .env.example\n```\n\n| Feature | Status |\n|---|---|\n| Multi-node federation | ✅ Live — node registry, manifest, federated query |\n| Server-side relationship generation | ✅ Live — `POST /relationship/generate` |\n| Node manifest | ✅ Live — `GET /.well-known/ions-node.json` |\n| Path registry | ✅ Live — `GET /path/{id}` |\n| Rights and attribution claims | 🔬 Experimental — framework in place, claims coming soon |\n| Token / reward mechanics | 🔜 Future |\n| Automated relationship generation on CBB approval | 🔜 Future |\n| Network-derived reputation scoring | 🔜 Future |\n| Multiple CBB types (observation, procedure, outcome) | 🔜 Future — claim type proven first |\n\n\"The model is less important than the network. Intelligence emerges through traversal and composition, not through parameter scale alone.\"\n\nIONS is built on the belief that:\n\n- Knowledge should be durable, inspectable, and composable\n- Reasoning should be visible, not opaque\n- Any model should be able to participate as an interpreter\n- Contributors should be able to see their knowledge being used\n\nThis is an open protocol. Fork it. Run a node. Contribute CBBs. Build on it.\n\nMIT\n\nPull requests welcome. For significant changes open an issue first to discuss what you'd like to change.\n\nThe most valuable contributions are:\n\n- High-quality CBBs in underrepresented domains\n- Improvements to the traversal engine\n- New relationship types with clear semantics\n- Documentation and examples", "url": "https://wpnews.pro/news/ions-a-distributed-reasoning-graph-built-from-evidence-backed-claims", "canonical_source": "https://github.com/nomad505050/ions-genesis", "published_at": "2026-06-24 03:07:09+00:00", "updated_at": "2026-06-24 03:14:03.739861+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "ai-infrastructure", "ai-tools", "large-language-models"], "entities": ["Ions", "Cognitive Building Blocks", "Genesis", "Claude Sonnet", "OpenRouter"], "alternates": {"html": "https://wpnews.pro/news/ions-a-distributed-reasoning-graph-built-from-evidence-backed-claims", "markdown": "https://wpnews.pro/news/ions-a-distributed-reasoning-graph-built-from-evidence-backed-claims.md", "text": "https://wpnews.pro/news/ions-a-distributed-reasoning-graph-built-from-evidence-backed-claims.txt", "jsonld": "https://wpnews.pro/news/ions-a-distributed-reasoning-graph-built-from-evidence-backed-claims.jsonld"}}