Action space. Every option a twin may choose from in a simulation. At least two, mutually exclusive, and complete — a missing option is redistributed across the ones you listed rather than being reported.
Alignment. How closely a population's answers have matched real people's answers on questions it had not seen. Measured continuously. See Validation.
Base population. Our model of the public, built from large-scale public studies and interviews and weighted to match the real composition of a market. Every population you define is drawn from it.
Combined. A label on part of an identity, meaning it was brought together from more than one study of the same kind of person. Reliable at the level of the pattern, looser at the level of the individual.
Comparison. Two or more simulations differing in exactly one field, read side by side, so the difference between them is attributable.
Confidence. The label on a result saying how far to trust it. Derived from how the population has performed against real people, the sample's resolving power, and whether the options actually separated.
Date. The moment a scenario is set in — not the day you run it. Determines which environment facts apply and what twins can know.
Digital twin. A simulated person: an AI agent given an identity built from real data about real people, placed in a situation, and asked to decide as themselves.
Drift. The change in a population's alignment since it was last tested. See Validation.
Environment. The facts every twin in a simulation treats as true. Dated, sourced, and shared across every twin in a run so that disagreements are about the people rather than about what each of them knew.
Estimated. A label on part of an identity, meaning it was assigned statistically because it was missing. Used to make a population's composition correct, and by default not part of what a twin is told about itself.
Groundedness. How much real evidence stands behind the people in a population. Reported alongside alignment: one is about the inputs, the other about the outputs.
Identity. Everything we know about one person and where each piece came from. Stable across every simulation that person appears in. Not a marketing persona.
Observed. A label on part of an identity, meaning it appears directly in a source.
Population. The specific group you ask, defined by you and drawn from the base population. Every simulation runs against exactly one.
Rationale. A recurring reason, grouped from the twins' own accounts of why they chose what they chose. Shown in a result with the twins behind it.
Reported. A label on part of an identity, meaning the person said it themselves.
Result. What a simulation gives back: the distribution, the ranges around it, the rationales, the subgroups, the confidence level, and every individual twin.
Run. One execution of a simulation definition, pinned to the exact versions of the population and environment it used. Running the same definition twice gives you two runs. See Simulation.
Sample. The subset of a population actually asked in a run, weighted to preserve the population's composition. A few thousand is normal.
Scenario. The situation each twin is placed in. Concrete and neutral; the setup rather than the question.
Simulation. One decision put to a group of twins, defined by five fields: population, date, scenario, question and action space.
Source. Where a piece of information came from. Every line of every identity and every environment fact carries one.
Subgroup. A slice of a population — by age, region, income — broken out separately in a result. Suppressed when too small to say anything about.
Takeaway. The one-line summary on a result. Written from the computed distribution and the rationales; never contains a number that was not counted.
Twin certainty. How sure a twin said it felt about its own choice. Self-reported, and not the same thing as confidence.
Version. A frozen state of a base population, a population, or an environment. Runs are pinned to versions so old results keep meaning what they meant.