causari-data

Open causal knowledge graph for AI agents and humans.

100+ curated events, 130+ causal links with evidence and confidence scores, 8 insight patterns — focused on the history of computing and software engineering.

This dataset powers @causari/mcp-server, a Model Context Protocol server that gives Claude Code, Cursor, and any MCP-compatible agent structured causal knowledge.

Data format

data/
├── events.json        # Historical events with metadata
├── links.json         # Causal relationships between events
└── insights.json      # Recurring causal patterns

packs/
└── worldcup-2026/     # Optional vertical pack for live causal timelines
    ├── README.md
    ├── events.json
    ├── links.json
    └── insights.json

Event schema

{
  "id": "kubernetes",
  "title": "Kubernetes",
  "description": "Google open-sourced its container orchestration system...",
  "yearNum": 2014,
  "yearLabel": "2014",
  "precision": "year",
  "domains": ["technology", "systems"],
  "impactScore": 0.82,
  "tags": ["containers", "orchestration", "kubernetes", "cloud"],
  "wikidataId": "Q22661306"
}

Causal link schema

{
  "id": "docker--caused-->kubernetes",
  "fromEvent": "docker",
  "toEvent": "kubernetes",
  "relationship": "caused",
  "confidence": 0.9,
  "evidence": "Kubernetes orchestrates Docker containers. Docker created the packaging format; K8s automated running them at scale."
}

Insight schema

{
  "id": "pattern--abstraction-layer-migration",
  "pattern": "Abstraction Layer Migration",
  "description": "Computing advances by adding abstraction layers that hide complexity...",
  "instances": ["fortran--enabled-->c_language", "..."],
  "predictiveValue": 0.82,
  "domains": ["technology"]
}

See SCHEMA.md for full field definitions.

Coverage

Vertical	Events	Links	Patterns
Computing & Software Engineering	50	81	3
AI / Machine Learning	15	15	—
Civilizational history	35	36	5
Total	100	132	8

Event packs

Event packs are optional vertical datasets for focused, faster-moving causal timelines.

They let Causari model domains where the key question is not just "what happened?" but:

what happened
→ why it mattered
→ who was affected
→ what changed next

Current packs:

Pack	Purpose
worldcup-2026	Proof of concept for live sports/event intelligence using World Cup 2026 results, causal implications, and watchpoints

See docs/PACKS.md for pack structure and quality guidelines.

Relationship types

Type	Meaning	Example
caused	A directly caused B	Printing press → Renaissance
enabled	A made B possible	TCP/IP → World Wide Web
accelerated	A sped up B	TypeScript → React adoption
inspired	A was conceptual seed for B	Unix → Linux
delayed	A slowed B	AI Winter → neural net research
prevented	A blocked B	(rare, used for counterfactuals)

Usage

With the MCP server (recommended)

{
  "mcpServers": {
    "causari": { "command": "npx", "args": ["-y", "@causari/mcp-server"] }
  }
}

Direct import

import events from '@causari/data/events.json';
import links from '@causari/data/links.json';
import insights from '@causari/data/insights.json';

Event pack import

For a live, daily-updated pack, fetch at runtime from a CDN so updates need no rebuild:

const BASE = 'https://raw.githubusercontent.com/causari/causari-data/main/packs/worldcup-2026';
const events = await fetch(`${BASE}/events.json`).then((r) => r.json());
// ...links.json, insights.json the same way

For a static pack, a build-time import is fine:

import worldCupEvents from '@causari/data/packs/worldcup-2026/events.json';

See docs/PACKS.md for consumption modes and docs/LIVE-UPDATES.md for the daily match-day workflow.

Raw JSON

Download from releases or clone the repo.

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Quick version:

1. Open an issue with the proposed-event or proposed-link label
2. Include: title, year, description, domains, and at least one source
3. For causal links: include evidence explaining why the causal relationship exists
4. A maintainer reviews and merges approved events in batches

Quality over quantity. Every event needs a clear description. Every link needs cited evidence. Confidence scores must be honest.

License

CC-BY-SA 4.0 — you can use, share, and adapt this data for any purpose (including commercial), as long as you give attribution and share derivatives under the same license.

See LICENSE for the full text.

Links

• @causari/mcp-server — MCP server for AI agents
• Powflow Canvas — visual explorer for humans
• causari.ai — project home