wealthschema/household-samples
WealthSchema Synthetic Household Samples 8 synthetic U.S. households, one per life stage plus one high-net-worth: a small free preview of what a complete, internally consistent household financial profile looks like. Each record covers the people, income, assets, debts, insurance, taxes, goals and a monthly trajectory. No real person is behind any of it. Built for teams that need realistic households to design, demo or test financial software: planning tools, robo-advisors… See the full description on the dataset page: https://huggingface.co/datasets/wealthschema/household-samples.
WealthSchema Synthetic Household Samples
8 synthetic U.S. households, one per life stage plus one high-net-worth: a small free preview of what a complete, internally consistent household financial profile looks like. Each record covers the people, income, assets, debts, insurance, taxes, goals and a monthly trajectory. No real person is behind any of it.
Built for teams that need realistic households to design, demo or test financial software: planning tools, robo-advisors, lending and insurance workflows, and AI agents for financial advice.
Why this exists
Testing wealth software usually means one of two bad options. Production client data needs a privacy review and a data-use agreement. Generated fake data (Faker and similar) breaks the moment your engine meets a realistic edge case, like a 28-year-old with a large brokerage account and no salary. These households are synthetic by construction, and their numbers are consistent with each other.
What's in a record
Load it:
from datasets import load_dataset
import json
ds = load_dataset("wealthschema/household-samples", split="train")
household = json.loads(ds[0]["household_json"])
print(ds[0]["archetype_name"], household["cash_flow"]["gross_income_annual"])What this sample leaves out, on purpose
This is a preview of shape and fidelity, not the product:
- Dollar amounts are banded (e.g.
$50k–$100k), not exact. - The monthly trajectory is truncated to 12 months; forward projections are withheld.
- Names, employers, metros and anything person-like are removed.
- Only 8 households: one per life stage plus one high-net-worth. The other 63 archetypes are in the paid corpus.
The full corpus has 1,451 households with exact figures, 96-month trajectories and Monte Carlo projections, sold as themed bundles or as a whole: wealthschema.com/datasets. A free watermarked single household (by email) is available at wealthschema.com/sample.
How the households are built
Deterministically, from seeded samplers: the same seed gives the same household. Demographics and balance sheets are calibrated to public sources, including the Federal Reserve's Survey of Consumer Finances. Tax and benefit figures are cited to IRS, SSA and CMS primary sources. Every record passes a schema gate before it ships. Schema overview: wealthschema.com/schemas.
License
Evaluation use, under the WealthSchema license. Every record carries _sample.license: "evaluation-only". Use it to evaluate fit, prototype and demo; don't redistribute it as a dataset of your own.
Related
- wealthschema/planning-benchmark: does an AI financial-advice system cite this year's figures?
- Rule Sets: current U.S. planning figures, each cited to its source, over an API and MCP server.
Questions: support@wealthschema.com
