About

Synthetic Users began by simulating individual research participants. We are now moving toward something harder: predicting quantitative outcomes at the scale of whole populations. Outcomes is the public record of that work — a blog that gives measurable predictions for real events by building and interviewing synthetic populations of the places those events concern.

Where the flagship product answers why one person thinks something, Outcomes asks how many, by how much, and which way — support and opposition, turnout and margin, sentiment and its spread across a community. The unit of analysis is the collective, and the output is a number with its reasoning attached.

What we predict

Anything where a population is about to decide or react: elections and by-elections, referendums, protests, contested local decisions, town halls, market and policy shifts — the general run of events that move through the world. If a real result will eventually land, it's fair game.

Why in public

Predicting how a whole population will decide or react is one of the hardest problems we know of. Human behavior is reflexive, context is everything, the ground truth arrives late and only once, and the space of things that can move an outcome is effectively unbounded. Nobody has this solved. A method that claims otherwise is either lucky or lying.

A problem this hard is better developed in the open than behind a wall. Private research can hide its misses, tune to the answer after the fact, and mistake a good story for a good model. Publishing forces the opposite discipline: we state the prediction before the event, state the method plainly, and grade ourselves against the real result when it lands — in public, misses included. That is the fastest way we know to find out what actually works, and the only honest way to earn trust in numbers that come from a simulation.

So Outcomes is a chronicle by design. It follows the development of quantitative prediction from synthetic populations as it happens — the techniques, the generative and other AI behind them, the wins and the failures — rather than a finished method presented as fact.

Disclaimer

Everything here is synthetic and experimental. The numbers describe simulated populations produced by our models — not polls of real people, not guarantees, and not forecasts you should rely on. Nothing on this site is financial, investment, legal, electoral, or professional advice of any kind. The underlying models are generative AI and can be wrong, biased, or hallucinate outright — stating something confidently is not the same as it being true. Treat every prediction as a research artifact, and make no decision on the basis of it.