Five fields define one. The quality of the answer depends almost entirely on how carefully you fill them in.
The five fields
| Field | What it is |
|---|---|
| Population | Who is being asked |
| Date | The moment the situation is set in |
| Scenario | The situation each twin is placed in |
| Question | What each twin is actually asked |
| Action space | Every option they may choose from |
A sixth thing shapes the answer without being a field: the environment, the facts every twin treats as true. It comes from your population and your date rather than from you, and you can inspect or correct it before running.
Definitions and runs
The five fields are a definition. Executing them is a run.
The distinction earns its keep the moment you do anything twice. You can run the same definition again — to check a result held, or after improving a population. You can change one field and run again, which is a comparison. Each run is pinned to the exact version of the population and environment it used, so an old run keeps meaning what it meant even after the population has moved on.
What happens when you run
A sample is drawn. A run does not use everyone in your population. It draws a sample weighted to preserve the population's composition, because a larger sample narrows the uncertainty with diminishing returns and rising cost. Two thousand is usually enough to separate options that are more than a few points apart.
Each twin decides on its own. Every sampled twin gets its identity, the environment at your date, and the decision in front of it. It reasons for itself and commits to one option. No twin sees any other twin's answer, and nothing is averaged into anyone.
The run finishes when every twin has decided. Then their individual commitments are added up.
This is worth being precise about, because it is what makes the product different from a forecast: the distribution is not estimated, it is counted. Nothing predicts that 63% will tip. Two thousand people each decide, and 63% of them tipped.
What it gives back
A result, which contains:
- the distribution across your action space, with a range around each share
- a confidence level saying how far to trust it
- the recurring reasons behind each option, with the twins who gave them
- how the answer differs across parts of your population
- every individual twin, readable and open to follow-up questions
How long it takes
Smaller simulations run in about a minute and you can watch the distribution form. Larger ones are queued and finish in the background, usually within the hour; you get the result when it is done rather than waiting on the page. Either way the run is durable — closing the tab does not affect it.
Where to spend your effort
In order of how much they move the answer:
- The population. Asking the right people beats everything else.
- The action space. A missing option corrupts every percentage.
- The scenario. Concrete situations produce sharper reasoning than topics.
- The date. Cheap to get wrong, expensive to notice.
- The question. Short, second person, about behaviour.
- The sample size. The least important lever on this list, and the one people reach for first.