Longterm Wiki

AI safety knowledge base — a wiki covering AI risk models, key organizations, researchers, policy approaches, and the AI Transition Model.

Built with Next.js 15 (App Router), MDX content, and YAML-driven data.

Getting Started

pnpm install
cd app && node scripts/build-data.mjs   # Generate database.json from YAML/MDX
pnpm dev                                 # Dev server on port 3001

Repository Structure

longterm-wiki/
├── content/docs/          # ~625 MDX wiki pages
├── data/                  # YAML data (entities, facts, resources, insights, graphs)
├── app/                   # Next.js 15 frontend
│   ├── src/               # React components, data layer, pages
│   ├── scripts/           # Build scripts (build-data.mjs)
│   └── public/            # Static assets + LLM files
└── crux/                  # Crux CLI — validation, analysis, content tools
    ├── crux.mjs           # CLI entry point
    ├── commands/          # 8 command domains
    ├── lib/               # Shared utilities + 34 validation rules
    └── validate/          # 23 validation scripts

Commands

All commands run from the repo root unless noted.

Build & Dev

pnpm dev                              # Start dev server (port 3001)
pnpm build                            # Production build (auto-runs build-data.mjs)
pnpm test                             # Run vitest tests

Crux CLI

The crux CLI provides validation, analysis, content management, and more.

# Validation
node crux/crux.mjs validate                  # Run all 28 validators
node crux/crux.mjs validate unified --fix    # Auto-fix with 34 rule engine

# Content
node crux/crux.mjs content create "topic"    # Create a new wiki page
node crux/crux.mjs content improve <page>    # AI-assisted page improvement
node crux/crux.mjs content regrade           # Re-grade quality ratings

# Analysis
node crux/crux.mjs analyze                   # Full health report
node crux/crux.mjs gaps list                 # Find pages needing more content
node crux/crux.mjs insights check            # Insight quality checks

# Auto-fixes
node crux/crux.mjs fix entity-links          # Convert markdown links to EntityLink
node crux/crux.mjs fix markdown              # Fix markdown formatting
node crux/crux.mjs fix escaping              # Fix LaTeX/JSX escaping issues

# Resources
node crux/crux.mjs resources list            # Pages with unconverted links
node crux/crux.mjs resources process <file>  # Convert links to <R> components

Run node crux/crux.mjs --help for the full command list.

Data Flow

data/*.yaml + content/docs/*.mdx
        ↓
  app/scripts/build-data.mjs       (build time)
        ↓
  app/src/data/database.json       (631 entities, 587 pages)
        ↓
  app/src/data/index.ts            (runtime data access)
        ↓
  Next.js static generation        (1300+ pre-rendered pages)

Entities come from two sources, merged at build time (YAML takes precedence):

1. YAML files in data/entities/ — rich entities with metadata, sources, related entries
2. MDX frontmatter — auto-creates minimal entities for pages with entityType field

Entity ID System

Entities have both string slugs (geoffrey-hinton) and numeric IDs (E42). Wiki pages use numeric IDs as canonical URLs. Each entity stores its numericId in its source file (YAML or MDX frontmatter); data/id-registry.json is a derived build artifact.

CI

GitHub Actions runs on push to main and PRs:

• build-and-test — builds data, runs 53 tests, builds Next.js (blocking)
• validate — runs the full validation suite (non-blocking, pre-existing content issues)