Research
A science of possible futures
Collective belief causes real-world outcomes: elections, runs, revolutions. We built the model class that makes it measurable, and we hold it to account in public.
The third foundational model class
Language models · 2018 →
Made language computable.
World models · 2024 →
Making physical reality computable.
Civilization models · 2026 →
Make collective belief computable. One entrant.
Every possible future already exists, distributed across eight billion minds. We measure which ones are likely.
One instrument. Three tenses.
Past · anchored
Decades of recorded human belief, carried in the civilization as memory.
Memory
Present · read
The live structure of any consensus: how much is conviction, how much is herding, where the hidden tension sits.
Structure
Future · mapped
Run the civilization forward, many times over, and read the distribution: futures that appear in every run are inevitable. Futures that appear in none are ruled out. The clusters are the basins history flows toward.
Distribution
We don't pick the future. We map the space of possible futures, with the impossible ones crossed out.
Held to account
Calibrated
Anchored to five decades of gold-standard human record: Pew, Gallup, and the World Values Survey, year by year since 1972.
Anchored
Benchmarked blind
Scored on questions the model has never seen, against outcomes that hadn't happened yet.
Blind
Published open
CIT-100, the first public benchmark for civilization models, scored in the open against live prediction markets, permanently.
In progress