The simulation
Seven steps, each with the population’s own numbers: how a question becomes a design, how the population is built and checked, how a sample answers one profile at a time, and what you get back. Every chart below is drawn live from the build in use.
A question in plain words becomes a design before anyone answers: the population it is asked of, the wording each person hears, the kind of answer and its options. You see the design with the answer, so nothing is asked that you did not mean.
Name the people in the question and the population follows: adults in Ohio, parents of teenagers, the self-employed, Republicans who voted in 2024. Two populations in one question become two runs, compared side by side.
Four finished runs and their designs: the population, the wording, the kind of answer, the sample.
“What would you pay a month for a streaming service?”
Every profile starts in the census. Each state's residents by single year of age, sex, race and Hispanic origin are reproduced cell by cell from the Census Bureau's 2025 estimates; national surveys add work, income, family, housing, health, civic life, party and the 2024 vote. Every other country and area is there too, by age and sex from the UN's estimates, shallower where the data is, and every answer says so.
The charts are the population's own counts, read live from the build in use.
Every resident of the United States in the population in use, by five-year band; the census, cell by cell.
One attribute from the survey record, read live for every state; the population is raked so each state matches its published table.
Open a family, then an attribute, to see its values and how the population splits across them; the tree goes deeper where the data does.
From
US Census Bureau population estimates, 2025
Every state by single year of age, sex, race and Hispanic origin, reproduced cell by cell.
From
American Community Survey 2024, calibrated to published tables
From
American Community Survey 2024, the person’s own household
From
American Community Survey 2024
From
National Health Interview Survey 2024 and the Behavioral Risk Factor Surveillance System 2024, by state
Drawn for every adult from survey distributions by sex, age, race and education, fitted to each state.
From
Current Population Survey supplements: voting and registration (November 2024), civic engagement and volunteering (September 2023); the Cooperative Election Study 2024 for party, ideology and the presidential vote
Drawn from each supplement’s distributions by state, sex and age band; the voting rates match the Census Bureau’s published figures.
Every country and area by age and sex, with fewer attributes than the United States.
From
United Nations World Population Prospects 2024
From
International Labour Organization and UNESCO statistics
From
World Bank and WHO country rates
A profile is not a list of independent traits. Each one is a real survey respondent's anonymised record, reweighted so that every state's totals match the published tables, so that age, work, income, family and health hold together the way they did in one life. No profile is a real person; every profile is built from one.
The curves show why that matters: employment by age for women and for men, with the shape the survey carries, because the joint was never taken apart.
A record from the population in use: the attributes it carries, and the persona written from them and nothing else.
As the model reads it
a 35-year-old woman living in Franklin County, Ohio; Asian; highest education: a bachelor's degree; occupation: registered nurses, in general medical and surgical hospitals, and specialty hospitals, at a nonprofit; personal income about $73,000 a year (from wages); married; born in Philippines, in the US since 2022, not a US citizen; speaks Tagalog at home, speaks English very well; insured through an employer plan; works about 36 hours a week; commutes about 10 minutes by car; degree in Nursing; household of 4 (with spouse and children) in a rented apartment; household income about $79,000; 2 vehicles; lives in the countryside; drinks alcohol; rates their own health as excellent; has been told they have blood sugar in the prediabetes range; is overweight; got no exercise in the past month; sleeps about 7 hours a night; a political independent; politically moderate; in the past year belongs to a group or association
The population's own counts. Two traits that move together with a third, kept together because every profile is one record.
Tables the build never used are predicted afterwards and scored, state by state, against what sampling alone would give. Attitudes are not attributes; they come from the simulation, and every kind of question is checked against real survey answers, item by item, within demographic groups. Ten open prediction markets were asked as expectation questions. Every number is public, misses included.
Where a kind of question has not been checked yet, the answer page says so in the sentence that gives the interval.
Each table the build never fitted, predicted afterwards and scored state by state; the median across the fifty states and the District of Columbia. A dot inside its band is a table reproduced as well as a sample of that size can be. Every check, with its number.
Each kind of question put to the population and compared with the 2024 General Social Survey and the 2024 Cooperative Election Study, item by item; the median distance in points.
10 open markets asked as expectation questions, three hundred profiles each; the population's yes among those with a view beside the price traders paid. The population sat above the market on 7 of 10.
A representative sample is drawn from the population. Each profile answers as itself, in batches of ten, and the estimate settles with its interval as more people are asked; the first hundred of every question are free. Then some of the people asked explain why in their own words, and the kinds of people in an answer are read out of the reasons.
“What would you pay a month for a streaming service?”: the running estimate after each batch of ten, with its 95% interval, over 2,000 profiles.
Two of the people asked, with their answers, from the same run.
No job, no room for subscriptions.
Seven dollars is about all I can spare from my monthly bills.
The kinds of people in this answer
23% to 40% of the sample of 120 · typically 7.08 US dollars per month
Who they are. They range from young adults to people in their eighties, with many not working or looking for work and describing very limited or fixed budgets.
What they say. They cannot spare money for another subscription, or can spare only a few dollars for entertainment.
“No room in my budget.”
Man, 26, Oregon · looking for work · 0
20% to 35% of the sample of 120 · typically 17.29 US dollars per month
Who they are. They are adults from their twenties through seventies who often watch after work or with a spouse, while still keeping an eye on other monthly bills.
What they say. They will pay around fifteen to twenty dollars because they use streaming frequently, but they do not want the bill to grow much larger.
“After work, streaming is how I unwind, and eighteen is reasonable for a service I use every day.”
Man, 56, Pennsylvania · 18
The free preview is the answer with its interval, who answered what, and why. A full run adds every breakdown the record allows, the states on a map, the market the answer implies, the kinds of people, every reason, and the next question asked of the same people; it prints as one document.
Four finished reports are public, so you can read what a full run looks like before asking for one.
The free sections and the ones a full run adds. Read a finished one: what people would pay a month for a streaming service, whether Ohio's self-employed would pay $50 a month for an AI bookkeeper, whether Ohio Republicans or Ohio Democrats would pay $10 a month for local news, what matters most when choosing where to buy groceries, ranked.
What this free preview shows, and what a full run adds.
The population knows what people report, not what they bought or clicked. Amounts are the weakest kind of question, where the model gives rounder numbers than people do. The build is a snapshot at 1 July 2025 and has not read this morning's news. Outside the United States the record is shallow. Attitudes are simulated and checked, never stored as facts about anyone.
Where this goes: a population you can follow, the same people asked again after the price changed or the policy passed; households that decide together; other countries with the depth the United States has now; a population grounded in what people do as well as what they say, from interviews and transactions, as that data arrives; and a population tested against launches, elections and markets, corrected where it is wrong.
Bring your data and get a population built for you alone.
Your customer lists, your surveys, your transactions or telemetry become a private population: matched to the same census frame, checked the same way, never mixed with anyone else’s and never part of the public population. Ask it what you would ask your customers, before you ask them.
Recruit a panel, prompt a model, or ask a calibrated population. They differ in what the answer is grounded in.
| A survey panel | Personas from a prompt | census.world | |
|---|---|---|---|
| Who answers | Recruited respondents, weighted afterwards | Characters a model invents from a description | One synthetic profile per resident, built from census and survey records |
| Representative of | The panel, after weighting | Whatever the prompt implies | The population, cell by cell: state, age, sex, race, origin, education, work, income, household, health |
| Sampling interval | Yes | No | Yes, on every number, narrowing as more profiles are asked |
| Checked against | Itself | Usually a handful of past surveys | Official tables held out of the build, real survey answers by demographic group, open markets |
| Reproducible | No: a new field, a new sample | No: a new prompt, a new crowd | Yes: versioned builds, the same person in every build of a lineage |
| Time | Weeks | Minutes | Minutes, with the interval shown as it forms |
Nothing here is a new idea. The population is built with methods statisticians have used for decades, applied end to end and checked at every step.