Traditional polling has a shape problem. You get a few hundred responses, weighted to look like a population, frozen at one moment in time. Ask a follow-up question and you have to field a new poll. Ask why and you get a checkbox.
A synthetic electorate is a different object. It is a population of simulated citizens — each with a demographic profile, a set of priors, and the capacity to answer open-ended questions — that you can query as often as you like.
What this unlocks
- Depth on demand. Every synthetic voter can be interviewed, not just tallied. “Would you vote for X?” is followed by “what would change your mind?”
- Counterfactuals. Rerun the same population under a different candidate, a different message, a different turnout assumption.
- Speed. A scenario that takes a polling firm two weeks and a five-figure budget runs in minutes.
What it does not do
A synthetic electorate is a model of a population, not a measurement of one. It inherits the biases of its construction. It is wrong in ways we are still mapping.
We treat every simulation as a hypothesis. The honest test is prediction: state the estimate before the real event, then compare. That is the entire point of publishing here in the open.
The loop we care about
1. Synthesize a population matched to real demographics
2. Pose the question — closed and open-ended
3. Aggregate, but keep the individual reasoning
4. When reality lands, score the simulation against it
5. Feed the error back into the next synthesisSteps 4 and 5 are the ones that keep us honest. Anyone can generate a confident number. The discipline is in grading it.
This blog is that loop, in public.