A Reproducible Birthplace Draw Experiment
We ran 20,000 seeded draws per mode through this game's actual country sampler. This tests the implementation against its own dataset; it does not validate the dataset or predict an individual life.
Method you can repeat
Dataset: 172 covered countries and territories. Each independent draw selects one country using annual births or population weights. Seed: birthplace-audit-v1. Source input snapshot SHA-256: dffaecfcf38da7e575f8d15f565a4c3d08cf8da7fe02fefb6bbe6468418e9c62. Run node scripts/model-experiment.mjs in the public repository. The script uses the actual chooseCountry function, not a separate replacement sampler.
Expected share and observed sample share
| Country | Mode | Model share | Sample share |
|---|---|---|---|
| India | Annual births | 17.599% | 17.390% |
| China | Annual births | 6.746% | 6.780% |
| Nigeria | Annual births | 5.692% | 5.645% |
| United States | Annual births | 2.772% | 2.730% |
| Japan | Annual births | 0.568% | 0.570% |
| South Korea | Annual births | 0.179% | 0.225% |
| India | Population | 17.801% | 17.520% |
| China | Population | 17.609% | 17.485% |
| Nigeria | Population | 2.821% | 2.825% |
| United States | Population | 4.252% | 4.020% |
| Japan | Population | 1.540% | 1.495% |
| South Korea | Population | 0.641% | 0.665% |
Checking the whole sample life
All 12 published language/country samples are rerun in a disposable browser. Validation compares a SHA-256 digest of age, character, complete state, complete event log, cause, death age and country, not age alone. The check is run against this engine; it is not independent demographic validation.
| Sample | Age | Event count | Outcome digest (prefix) |
|---|---|---|---|
| JPN | 98 | 99 | 776abf90de9e36da |
| USA | 68 | 68 | e0d0d881eb39a369 |
| IND | 82 | 83 | 3ef4f8fe0b3e80d1 |
All sample proof records (JSON). Run node tests/release-browser.cjs against the local preview. Text/localization changes can change the digest; published seed rules and numeric weights are separately versioned.
What the result means
Finite random samples differ from exact model weights. Repeating this exact seed and version repeats these counts; choosing a different seed changes them. Agreement with the model only checks internal behavior. The experiment does not establish real-world accuracy, demographic fairness, mortality calibration or user outcomes.
Observation years and primary-source provenance for the inherited numeric fields remain unverified. Results reviewed 20 October 2026. Download results (JSON) · Sources and limitations · Try country comparisons