Human Knowledge Map

A zoomable, explorable map of human knowledge — from primary-school concepts to research frontiers.

Vision

Human Knowledge Map treats knowledge as a graph rather than a list of courses. The product combines three layers:

1. Human Knowledge Graph — what humanity knows.
2. Curriculum Graph — good ways to learn it.
3. Personal Knowledge Graph — what a learner already knows, is learning, or is likely to forget.

Current version: v7.2

Explorer and learning layer

• v1.3 Semantic Zoom
• v1.4 Domain Intelligence
• v1.5 Learner State
• v1.6 Forgetting Model
• v1.7 Recommendation Engine
• v1.8 Provenance
• v1.9 Graph Validation
• v2.0 Graph I/O
• v2.1 AI Expansion
• v2.2 Adaptive Explorer

Knowledge data pipeline

• v2.3 Ontology Decomposition
• v2.4 Deduplication
• v2.5 Quality Scoring
• v2.6 Import Staging
• v2.7 Source Trust Policy
• v2.8 Growth Planner
• v2.9 Domain Bridges
• v3.0 Curriculum Ingestion
• v3.1 Graph Metrics
• v3.2 Dataset Registry

Large-graph architecture

• v3.3 Math Foundation Dataset
• v3.4 Graph Repository
• v3.5 Spatial Index
• v3.6 LOD Clustering
• v3.7 Graph Query API
• v3.8 Graph Chunks
• v3.9 Search Index
• v4.0 Search API
• v4.1 Stress Graph Generator
• v4.2 Large Graph Lab

Production data layer

• v4.3 PostgreSQL Schema
• v4.4 Graph Change Sets
• v4.5 Version History
• v4.6 Review Workflow
• v4.7 Ingestion Jobs
• v4.8 Math Curriculum Catalog
• v4.9 Canonical Merge Engine
• v5.0 Review Queue API
• v5.1 Graph Event Log
• v5.2 PostgreSQL Repository Adapter

Knowledge operations layer

• v5.3 Admin Console
• v5.4 Repository Factory
• v5.5 Batch Expansion Planner
• v5.6 Import Preview
• v5.7 Publish Pipeline
• v5.8 Expanded Math Curriculum Scaffold
• v5.9 Operations Metrics
• v6.0 Admin API
• v6.1 Release Manifest
• v6.2 Integrated Operations Dashboard

Maintenance and editorial layer

• v6.3 Database Migration Registry — ordered migration metadata and review/history SQL migration.
• v6.4 Canonical Seed Planner — builds safe, reproducible seed batches before persistence.
• v6.5 Review Diff Engine — converts change-set mutations into reviewer-friendly diffs and counts.
• v6.6 Rollback Planner — creates inverse change sets from pre-publication snapshots.
• v6.7 Review Workspace — /admin/review provides an editorial workspace for inspecting queued change sets and notes.
• v6.8 Controlled Generation Console — /admin/generate exposes bounded AI proposal generation without bypassing review.
• v6.9 Provenance Coverage — measures sourced, reviewed and verified coverage and identifies missing provenance.
• v7.0 Canonical Math Seed — mathematics nodes are packaged with explicit provenance/review status; internal editorial review is not treated as external certification.
• v7.1 Publish / Rollback API — POST /api/publish returns release manifests or rollback plans while keeping canonical mutation behind the repository transaction layer.
• v7.2 Knowledge Quality Dashboard — /admin/quality surfaces provenance coverage and missing-source work.

The central learning path remains easy to inspect:

计数 → 自然数 → 加法 → 乘法 → 分数 → 代数 → 函数 → 微积分 / 线性代数 / 概率 → 神经网络 → Attention → Transformer → FlashAttention

Knowledge lifecycle

Curriculum / textbook / paper / expert / AI proposal
                    ↓
            Controlled expansion batch
                    ↓
               Ingestion Job
                    ↓
       Import preview / normalize / dedupe
                    ↓
             Canonical Merge
                    ↓
              Change Set
                    ↓
              Review Diff
                    ↓
             Editorial Review
                    ↓
              Publish Plan
                    ↓
             Release Manifest
                    ↓
         PostgreSQL Knowledge Store
                    ↓
             Graph Event Log
                    ↓
 Graph API / Search / LOD / viewport clients

AI-generated content never becomes canonical merely because a model produced it. Curriculum scaffold data is also treated as a proposal until mapped to sources and reviewed. reviewed means internal editorial review; stronger verified status is reserved for a stricter external-source verification workflow.

Database and runtime storage

The initial PostgreSQL schema is in db/schema.sql; later migrations are tracked in db/migrations/ and src/lib/migrations.ts.

GRAPH_STORE=memory
# or
GRAPH_STORE=postgres
DATABASE_URL=postgres://...
DATABASE_POOL_SIZE=10
DATABASE_STATEMENT_TIMEOUT_MS=5000

Run locally

npm install
npm run dev

Open:

• http://localhost:3000 — adaptive knowledge explorer
• http://localhost:3000/lab — progressive large-graph lab
• http://localhost:3000/admin — knowledge operations console
• http://localhost:3000/admin/review — change-set review workspace
• http://localhost:3000/admin/generate — controlled AI generation console
• http://localhost:3000/admin/quality — provenance and editorial quality dashboard
• http://localhost:3000/api/admin — operational summary API
• POST /api/publish — release or rollback planning API

Type-check:

npm run typecheck

Generate a 100k-node stress dataset:

npm run stress:graph -- 100000 ./stress-graph.json

Docker deployment

The application image is based on Debian Bookworm and configures the default APT source to:

https://mirrors.ustc.edu.cn/debian

The default Compose stack runs the Next.js application together with PostgreSQL 17 (Debian Bookworm).

Prepare environment variables:

cp .env.example .env
# edit POSTGRES_PASSWORD and DEEPSEEK_API_KEY

Start the full stack:

sh scripts/deploy.sh up

Useful commands:

sh scripts/deploy.sh status
sh scripts/deploy.sh logs
sh scripts/deploy.sh restart
sh scripts/deploy.sh pull
sh scripts/deploy.sh down

Apply database schema and migrations manually:

sh scripts/migrate.sh

The legacy convenience command also starts the Compose stack:

sh scripts/start.sh

By default the service listens on port 3000. Override it in .env with PORT=8080 or another port.

reset-db deletes the PostgreSQL Docker volume and all persisted knowledge data, and therefore requires typing RESET interactively:

sh scripts/deploy.sh reset-db

DeepSeek

Copy .env.example to .env.local for a non-Docker local environment and set your key. POST /api/expand proposes AI-generated graph expansions. AI output must pass staging, provenance, deduplication, quality scoring, graph validation and review before publication.

Stack

• Next.js
• React
• TypeScript
• React Flow
• PostgreSQL-backed repository layer
• Docker / Docker Compose
• DeepSeek API (optional graph expansion)

See docs/ARCHITECTURE.md and docs/ROADMAP.md for product and technical direction.