The Popularity Graph
4ort.xyz is a knowledge graph of every entity in Wikidata — people, places, software, films, companies, ideas — ranked by real-world attention. Each entity gets a popularity score derived from Wikipedia pageviews, citation count, and edit velocity. Every claim is sourced. Built for AI agents, search engines, and humans who want the world’s facts with citations.
What makes it different
Wikipedia tells you what exists. 4ort.xyz tells you which one matters most, why, and by how much. We do this by ingesting:
- The full Wikidata dump (152 GB), with every claim, qualifier, and citation preserved
- Hourly Wikipedia pageview data from the Wikimedia clickstream
- Live edit firehose from Wikidata EventStreams (95% of edits applied within minutes)
- SEC EDGAR filings for public companies
- Federal economic data from BLS, BEA, Census, EIA, Treasury, and the Federal Reserve Board
Then we cross-rank. The result: an entity-by-entity scoreboard of human attention, refreshed continuously, citable by AI.
How the popularity score works
Every entity carries one number from 0 to 100. It blends three signals:
- 35% — Wiki Languages. How many Wikipedia language editions have an article about it. Breadth of recognition.
- 35% — Unique Citations. How many distinct sources back its facts. Depth of the record.
- 30% — Wikipedia Views. How many people actually read about it in a typical month.
Views are log-scaled against the 99.9th percentile of every entity we hold views for — 13,409 views a month, measured from the data, not chosen by hand. An entity at or above that reads as full marks for attention; everything below is spread across the scale. We publish the percentile each score was measured against alongside the score itself, so the number always says what it was compared to.
Where a language edition’s current month is missing, we use that entity’s typical month (its year divided by twelve) rather than reading the gap as “nobody read this”. It keeps the unit honest and stops seven languages’ worth of subjects from scoring zero on attention for a reason that has nothing to do with them.
Operating principles
How we make money
Right now: nothing — the site is free and ad-free. Soon: a paid API tier for high-volume commercial use (AI startups, ML teams, journalism tools). The website itself stays free, fast, and ad-free forever.
Technology stack
Python + FastAPI + PostgreSQL with pgvector + Redis, hosted on a single Hetzner VPS in Germany. Server-rendered HTML for speed and SEO. The Wikidata dump is processed locally; live updates via the EventStreams SSE firehose. CDN via Cloudflare; analytics via self-hosted Umami (cookie-free). No third-party tracking scripts.