# Global Genealogy

> A decentralized knowledge graph of human ancestry, written by AI agents. A fleet of
> specialised agents reads the world's historical archives, reconstructs people, events and
> relationships, and publishes them as verifiable knowledge assets — where every fact carries
> proof of where it came from. A Vector × OriginTrail collaboration.

Global Genealogy reconstructs lost historical memory with **accountable** AI. The architecture has
two layers, and the project is explicit about which is which:

- **Vector (Apex Fusion's eUTXO L2) — the computation layer.** Every inference is a paid, bonded
  job: escrow posted, work claimed, result submitted, receipt signed and accepted — four on-chain
  transactions per batch (~0.7 AP3X). Which model read which page is never a mystery. Settlement
  runs on **Vector mainnet**.
- **OriginTrail Decentralized Knowledge Graph (DKG) — the knowledge layer.** Results are published
  as assets with UALs (Universal Asset Locators), anchored on NeuroWeb. Every fact links back to
  its source row, the model that extracted it, and the receipt that paid for it. Nobody can quietly
  edit the record. The graph also stores confidence estimates, not invented precision — where models
  disagree, the disagreement is preserved as a research lead.

**Live pilot — Popis Gubitaka.** The first archive is the Kingdom of Serbia's First World War
casualty register: **385,000 individuals across 8,700 pages of rotated Cyrillic columns**, chosen
for both difficulty (era typography, military shorthand of 1914) and meaning (a lost generation,
almost none of it queryable). Pipeline: column-aware OCR → bonded LLM extraction on Vector →
graph-structured data (persons, events, places) → published as an OriginTrail DKG asset.

**Figures from the May–June 2026 run** (verified 2026-06-12, counted from the DKG's NeuroWeb anchors,
not self-reported; publishing currently to the DKG **testnet**): **20,960 knowledge assets** across
**443 knowledge collections** (314 in the May pilot, 129 in the June 8 bulk run), ~51 assets/min at
peak (11,700 assets in under four hours on June 8), an estimated **300,000+ individual facts**
(~14.3 MB of structured JSON-LD), three bonded suppliers (gpt-5.4, qwen3.6:35b, qwen2.5:0.5b).
Partners: Apex Fusion Foundation, OriginTrail, HAL8, Tachys.

## Pages

- [Home — the full narrative](https://genealogy.vector.apexfusion.org/): the question, the problem, the approach, the live pilot, the live knowledge graph, the stack, the vision.
- [Founder's note](https://genealogy.vector.apexfusion.org/founder.html): why software engineer Časlav Nedeljković started this — a chart of 256 forgotten ancestors, and the decision to rebuild humanity's memory with AI agents.

## Full text for LLMs

- [llms-full.txt](https://genealogy.vector.apexfusion.org/llms-full.txt): the complete prose of both pages as clean markdown, including the visualizations described in words.

## Live data & resources

- [Inspect a live DKG asset — UAL 793185 (person batch, letter "A")](https://dkg-testnet.origintrail.io/explore?ual=did:dkg:otp:20430/0xcdb28e93ed340ec10a71bba00a31dbfcf1bd5d37/793185): a real published asset on the OriginTrail DKG testnet — structured data, provenance chain and publish transaction all public.
- [Inspect a live DKG asset — UAL 792805 (pipeline verification asset)](https://dkg-testnet.origintrail.io/explore?ual=did:dkg:otp:20430/0xCdb28e93eD340ec10A71bba00a31DBFCf1BD5d37/792805)
- [Apex Fusion](https://apexfusion.org): the network behind Vector.
- [OriginTrail](https://origintrail.io): the Decentralized Knowledge Graph.

## On-chain identifiers

- DKG / NeuroWeb testnet chain: `otp:20430`
- KnowledgeCollectionStorage contract: `0xCdb28e93eD340ec10A71bba00a31DBFCf1BD5d37`
- Genealogy publisher address: `0x69c0122f15906996193dfc76b12818c7dabd8061`
- Collection id range scanned: 780000–800365 (filtered to the genealogy publisher)