An autonomous AI system explored Wikipedia with Monte Carlo Tree Search and discovered real, verified coincidences across history. Every fact is sourced. Every connection is true.
The simulation is showing seamsRanked by improbability, verifiability, and narrative satisfaction. Scored by Claude.
A 7-level autocatalytic system. Each level feeds into the others.
The worst attempts. So forced they're funny for different reasons.
The short version for impatient humans.
An AI downloaded all of English Wikipedia (6.9 million articles) into a local database. Then it used Monte Carlo Tree Search — the same algorithm that beat humans at Go — to systematically explore connections between articles.
At each step, it picks a strategy: follow a shared date, find a name collision, trace an etymology, find a geographic mirror, or one of 8 other approaches. It builds chains of 3–6 articles, then scores them on verifiability (are the facts real?), improbability (how unlikely?), and narrative satisfaction (does the universe seem to have a sense of humor?).
The system improves itself as it runs: a genetic algorithm evolves better search strategies, a meta-analysis engine discovers which types of coincidences are richest, and a Red Team tries to debunk every discovery. Only the ones that survive get published here.
Everything on this page is real, sourced, and verified. Click any Wikipedia link to check for yourself.