This is a toy historical model, not a precise dataset. Do not cite it. Thirty-eight world regions, plus twenty-one city-scale and diaspora overlays, each carry hand-curated anchor estimates at 19 moments between AD 33 and 2026, linearly interpolated between them. Four datasets ship: the Christian, Muslim and Buddhist share of local population, and the share living in settlements of 5,000 or more.
The sources are standard scholarship rather than any single authority. Rodney Stark's early-growth estimates (roughly 40% per decade to AD 300), dated conversions (Armenia 301, Ethiopia c. 340, Rus 988), post-conquest decline curves, colonial-era spread, and modern Pew and World Religion Database figures for the faiths; Bairoch, de Vries, Malanima, Anthony Reid, McEvedy & Jones and UN World Urbanization Prospects for population and urbanisation. Where they disagree the middle was taken and the caveat widened, rather than the number narrowed. The README says which figures a knowledgeable reader should challenge, and why.
Confidence is drawn, not footnoted. Fills wash toward grey in proportion to how much is actually known: a figure reconstructed from monastic records does not render at the same strength as one from a national census, and the tooltip names the tier. Some quantities are contested and do not improve with time — France bans religious censuses, and "Buddhist" is not a well-defined share in Vietnam, where the 2019 state census returns 4.8% and a 2024 survey returns 38% of the same people.
Faiths are mutually exclusive shares of one population, so they contest each other for colour; with two visible, ground held by both blends, and with three it is striped, because the blend between two hues occupies the space a third would need. Urbanisation is a measure rather than a faith: a population can be 90% Christian and 90% urban at once, so it fills the map on its own ramp while the faiths reduce to their frontier lines over it.
The simulation layer is a logistic growth plus diffusion model on a 2° land grid, one pass per faith, seeded at Jerusalem in AD 33, at Mecca and Medina in 622, and at Magadha, Gandhara and Anuradhapura for Buddhism, which was already five centuries old at the first anchor. It spreads along populated cells, weakly across narrow seas, and over dated ocean and overland routes. Decline eras are read out of drops in the anchor data, so each faith feels its own collapses. The blend slider mixes the two layers; at 100% history you see the anchor data, at 0% the model runs free under your parameters.
Heat intensity is the log-scaled share multiplied by a population-density weight, so sparse deserts stay dark even where a faith is dominant and early minority communities remain visible. The render field is an analytic gaussian with no land mask, which means a small region beside a huge one is diluted by its neighbour's population mass: Sri Lanka renders well below its true Buddhist share because India carries 68 million people against its 1.2. The leading dataset is still correct; only the intensity understates. Click a region to see its real curve.