Population Scenario Visualizer

Live at https://alejandrozu.github.io/population-visualizer/ (the guide is at #explanation).

To run it locally, serve the folder with any static web server, for example python -m http.server 8766, and open http://127.0.0.1:8766/index.html.

The project explores two claims about population under longevity escape velocity (LEV):

1. With fertility well below replacement, population approaches a ceiling even if nobody ever dies: roughly 1.5 to 3 times today's population for fertility between about 1 and 1.6 children per woman. Fertility sets the height of the ceiling; longevity decides whether there is a ceiling or a decline.
2. Under LEV, a population with a very high median age is not old in health terms. The feel-age measures make this visible.

What it does

• Provides two top-level views: the interactive Simulator and an Explanation that sets out the argument, the rationale for every lever and preset, the methodology, the original concepts (feel age, survival-adjusted TFR, caution), sources and limitations. The explanation can be linked directly with #explanation.
• Loads public demographic CSVs live from Our World in Data grapher endpoints.
• Covers world, regions, and countries where the source series exist.
• Uses historical estimates through 2022. Years 2023 onward are treated as simulation/projection years because recent source columns are inconsistent across places.
• Computes births, deaths, TFR, modeled under-15-survival-adjusted TFR, population, percentage growth, absolute growth, absolute migration, migration per mille, life expectancy, selected age share, median/percentile age, health-adjusted percentile age, a normal population pyramid, and a feel-age comparison pyramid.
• Simulates from 2022 onward with editable and reorderable checkpoints for TFR, life expectancy, net migration per mille, and a caution factor that subtracts deaths per 1,000 from the mortality curve.
• Uses the actual 2022 value as the baseline for forward interpolation, then interpolates toward later checkpoints.
• Default displayed range is 1950-2100, with 2022 as the historical anchor and 2023 onward simulated.
• Age/mortality arrays are finite-capped at 500. Pyramid displays use the age where mortality reaches 990 per 1,000 when it occurs earlier.
• Mortality curves start with LE 75 and the selected comparison year. Additional year or life-expectancy curves can be added inside the mortality panel, up to 10 total.
• The 2022 simulation handoff is anchored to observed births, deaths, TFR, and life expectancy before simulated years continue forward.
• Annual birth counts are preferred whenever present (the lower of the two loaded series); crude-rate/model estimates are used only as fallbacks.
• Age-share, pyramid, feel-age comparison, and health-age controls live inside their own panels.
• Historical age-share and percentile-age rows use annual cohort distributions when age-group data is available, so the 2022 simulation handoff does not switch from a modeled age share to a different cohort method.
• The feel-age comparison pyramid scales each half independently while keeping hover values in real people counts.
• Population pyramid panels accept historical years before 2022 and use horizontal age hover behavior.
• Default panels show population, births/deaths, TFR, life expectancy, absolute growth, mortality, and median/percentile age; growth rate, migration, pyramids, age share, and health/feel age start hidden.
• Supports staged editing with Dynamic off and manual recomputation with Update. Chart-only settings (Y starts at 0, pyramid year and height, age-share range, percentile, mortality curves) redraw only their own panel and apply immediately.
• The Simulation tab shows 1950-2100; the History tab shows only the observed record, 1950-2022. Switching tabs keeps the panel layout.
• Reset restores the default place, ranges, checkpoints, panels, panel controls, and curve selections. Export downloads the current scenario as JSON.
• Export also copies the scenario JSON to the clipboard when the browser blocks direct downloads.
• Major line charts include an optional Y starts at 0 checkbox inside the panel.
• Separates generalized mortality curves from disease-only curves for feel-age calculations.

Important assumptions

• Pre-1950 births are estimated from long-run birth rates and population where direct annual birth counts are unavailable.
• Pre-1950 deaths are derived from births minus annual population change when direct annual death counts are unavailable, with a modeled mortality fallback from life expectancy.
• Pre-1950 TFR is inferred from births, population, growth, and the modeled survival schedule, then smoothly calibrated to the first observed TFR for that place.
• Under-15 survival-adjusted TFR is derived from the generalized mortality curve implied by that place's life expectancy in that year. For recent decades that curve overstates child mortality (about 9% before 15 for the world in 2022, against about 4.6% observed).
• Historical net migration is a residual: population change minus births plus deaths. Forward migration uses the checkpoint per-mille assumption.
• World historical migration is forced to zero. Forward world migration remains adjustable for hypothetical scenarios.
• Five-year age groups are smoothed into single-year ages with interpolation before pyramid rendering.
• Mortality curves beyond observed modern life expectancies are extrapolations. They are useful for scenario exploration, not forecasts.
• Where two loaded birth or fertility series disagree, the lower is used, because UN estimates have tended to run above registered counts. It does not scrape Wikipedia tables automatically.
• Life-expectancy presets are annual gains from each place's own 2022 value and are recomputed when the place changes, unless the life checkpoints were edited by hand. LEV presets keep their stated gain until the model ceiling of 500 (a checkpoint is added in the year it is reached); in practice the curve stops improving near 456 because of the background death rate, which caution removes.
• Fertility presets are world paths applied unchanged to every place.

Performance notes

• Historical getters are memoized per place and year, and interpolation uses sorted per-series indexes instead of re-sorting rows on every lookup.
• Mortality curves are fitted with a precomputed Gompertz age factor, and feel-age maps are computed once per life expectancy.
• The forward simulation is cached per place and checkpoint set and extended on demand (for example, for a pyramid year beyond the chart window).
• Three datasets that were downloaded but never used (under-15 mortality, death rates by age, growth rate; about 2.9 MB) are no longer loaded.
• Renders are coalesced into one animation frame, typing is debounced, and color-banded lines use one Plotly trace per color instead of one per year.