About this Atlas

How to read this map

This Atlas is a finding aid, not a citable source. Use it to explore when and where things happened. Every ruler, event, and article links back to Wikidata or Wikipedia, where you can follow the sources. A few limits to keep in mind as you explore:

• All borders are approximate. Many historical boundaries were zones, frontiers, or networks of tribute rather than lines. Crisp polygons impose a modern idea of territory on steppe empires, maritime trade powers, and tributary systems that never worked that way. That is a limit of the medium, and it applies to every shape you see here.
• Shapes between keyframes are interpolations. Border geometry is hand-drawn at coarse year steps and morphed smoothly between them. The in-between frames are animation, not historical claims. The steps are uneven (centuries wide in the ancient world, decades in the modern), and even the modern record has holes: it jumps from 1815 straight to 1880, so whole famous episodes, like the unification of Germany (1871) or of Italy (1861–70), fall between frames and never appear. The panel dates, not the morphing shapes, carry the real chronology.
• A fading border means the date is soft. When a polity ends, its outline dissolves instead of vanishing. Where the end is a date somebody wrote down (a treaty, a dissolution, an abolition), it dissolves over about 25 years. Where the same polity returns to the map soon after, the fade lasts until its next chapter starts fading in, so the two meet instead of leaving a gap. Where all we really know is that our border source stops, it fades for roughly a lifetime instead: that is the case for most peoples and for much of the pre-modern world, and the long fade is a limit of the data, not a claim that anyone disappeared. And where nothing else ever claims the same ground, the outline does not go out at all: it holds as a faint trace, for centuries if the map stays empty there, and yields only when a later polity takes the space. A blank map would say “no one was here”, and that is rarely true. Births mirror this: where the founding is a recorded date, the outline arrives within a few years; where our border record simply begins, it fades in over a lifetime, marking the start of our data rather than the birth of a people.
• Recent borders use two sources. The historical border record ends at 2010. Where a country's borders have actually changed since then (South Sudan's independence, Crimea), the map carries a present-day frame derived from Natural Earth's reference data, and Natural Earth's conventions decide the contested cases: it maps de facto control, so Crimea appears under Russian control, and de facto states it maps — Kosovo, Palestine, Northern Cyprus, Somaliland — appear here too. Countries whose borders have not changed keep their own most recent frame, which draws the same line in more detail. In frames from 1990 onward, islands the historical record never drew (outlying archipelagos, fjord islands, the high Arctic) are filled in from Natural Earth so they appear on the modern map. Showing a source's lines is not an endorsement of any boundary or claim, there or anywhere on this map. A couple dozen small territories and microstates absent from that dataset still hold their 2010 shapes.
• The Americas at 1492 show traditional indigenous territories. For the moment of European contact, the border source incorporates territory mappings that trace back to Native Land Digital — including names written in their own orthographies, like Očhéthi Šakówiŋ and Anishinabewaki. These depict traditional territories: asserted, often overlapping, and not legal boundaries; the peoples shown lived there long before 1492 and most still do. The map holds this layer from 1492 until the next border sheet at 1500, and we left out groups whose formation came after contact (the Métis, for example) rather than draw them anachronistically. Some smaller territories and a few peoples without usable records are still missing — tracked for future refreshes. The same applies to the indigenous territories shown elsewhere (Australia, the Pacific): boundaries are indicative, some are contested and may not be agreed to by all traditional custodians, and nothing here is suitable for native title or land claims — a caveat we borrow from the AIATSIS map of Indigenous Australia, whose principles guide this layer. Where a community's preferred spelling differs from our map sheet's register, the panel carries the preferred form (checked against AUSTLANG, the AIATSIS language database). And a date range on a people's panel marks the span our map sources cover — never an end: these are continuing peoples, most living in their territories today.
• The faint modern country borders are a deliberate anachronism: an orientation aid, so you always know where on today's Earth you are looking. They imply no endorsement of any present-day boundary, and neither do the historical shapes.
• One religion label per era is a massive simplification. Every society in history was religiously plural. The religion view paints each polity with a single family-level color per era — a coarse index, not a description of belief.
• Some source entries are combined; others are split by era. A polity the border atlas renames across eras (United States of America → United States) is unified into one entry; one whose shape vanishes for centuries (Egypt under Persian, Roman, and Ottoman rule) splits into separate era entries, each honest about its own span. These joins are editorial choices, disclosed here, not statements of fact.
• “In their own language” is a polity's own name for itself. Where our sources record one, the panel carries the name a state or people used in its own language and script — Ελλάδα Elláda for Greece, საქართველო Sakartvelo for medieval Georgia — with a romanization when the script needs one. The line appears only when that name differs from the display name, and it reflects the era our sources cover, not every name a community uses today.
• Polity colors carry no historical meaning. The palette is presentational and re-assigned at load time purely to keep neighbors distinguishable.
• Era names are conventions. Scrubber epochs like “Early Modern” are Western periodization applied globally — a labeling convenience.

How it's built

The map and its historical data

The Atlas has two kinds of content. The geometry layer (which polities exist, when, and what shape they hold) is curated for this project. The knowledge layer — rulers, events, notable people, article summaries — comes from Wikidata and Wikipedia, gathered and verified in a snapshot assembled over several fetch days and shipped inside the app itself, so every click answers instantly, offline included. Data as of 2026-08-25. Reader-card summaries still refresh live from Wikipedia when you're online, so quoted prose stays current. When the timeline reaches its final year the masthead and the browser tab read The World Now — a nickname for the end of the timeline, not a live feed: the last recorded borders and offices are held forward until the data is refreshed.

What we curated by hand

• The polity roster and lifespans. Displayed dates prefer Wikidata's inception/dissolution claims when they agree with the map's coverage; where geometry coverage differs from the historical dates, the polity's panel shows an “on this map …” footnote disclosing the gap.
• Chapter dates. Where one entry spans several political eras (shogunates, empires, republics), the transition years shown on hover are individually sourced, not read off the map sheets.
• Border keyframes, derived from the open historical-basemaps project (A. Ourednik): year-snapshot world boundaries, simplified and composited for this map. The source describes these boundaries as approximate, and the same caveat applies here.
• Religion breakpoints and culture tags, coarse and family-level by design. A core set is hand-curated with era breakpoints; the rest is seeded from each polity's Wikidata claims. Where those claims carry dates, the seed is era-aware (Makuria turns Christian in 600 and Islamic after 1317, the Golden Horde converts under Uzbeg Khan, Scandinavia's Reformation is dated), and a hand-curated list carries famous transitions the data omits. Most polities still show a single value across their whole mapped span, which flattens history (a state's current official religion is what Wikidata records). The value is the religion of the state (the court, crown, or ruling institution), not a population census: for earlier polities that means the state cult or established church, and the breakpoints are court events (conversions, establishments, disestablishments). For today's states the paint follows what each government itself says: an established or official religion keeps its color, a state that has declared itself secular or formally ended its state church paints as “None (secular state)” from that date, and a state that declares neither shows as unrecorded — a majority faith is never painted as if it were official.
• The religion view groups labels into families, with the major branches of the largest traditions as families of their own: Catholic, Orthodox and Protestant Christianity (the Reformation and the East–West divide are visible as color), Sunni and Shia Islam (the Ottoman–Safavid border), and Theravada and Tibetan Buddhism. Records that only say “Christianity,” “Islam,” or “Buddhism” paint as the undivided tradition, which is also how the early centuries of each faith are deliberately shown. Finer distinctions (Lutheran vs Calvinist, Eastern vs Oriental Orthodox, Ibadi Islam) change the details panel, not the paint.
• Language breakpoints, likewise coarse: hand-curated for the core set, seeded from Wikidata's official-language claims for the rest (era-aware where the claims are dated — León's chancery switches from Latin to Spanish in 1230 — single-value otherwise, disclosed here). The language map mode colors each polity by its principal administrative or court language, never a claim about everything its people spoke; every empire on this map was multilingual. Painting is by language family; Indo-European and Afro-Asiatic are each divided one branch further, following the convention of linguistic atlases. Rome and the Spanish Empire share a color, for example, because their administrative languages are related. Where a state's language is unrecorded (Tiwanaku, Wari, Chimú), the map says “unrecorded” rather than guessing.
• A few ruler chronologies. Reigns come from Wikidata, but it sometimes knows a ruler and not when they ruled, and for a handful of polities it has no ruler list at all. There we supply the conventional dates from the standard published chronology, and any dates Wikidata later gains take precedence over ours. Three polities needed more extensive manual chronologies because Wikidata does not identify the relevant ruling offices. The Kingdom of Bosnia's kings hold no recognized office, so six of its ten cards are transcribed by hand. The two Emilian duchies, Modena and Parma, have none of their rulers recorded as such: every office the data does scope to them is a family style or an heir's title, so both ducal lines, 27 rulers between them, are hand transcriptions of the standard regnal tables. Each card is the name, the title, the reign and the link, checked against that ruler's own dates before it shipped. The cards transcribe the published chronologies, and each links out so you can check the entry.
• Identity resolution. Each polity is matched to its Wikidata entity by hand, with every match verified live against the entity's labels and descriptions. With more than a thousand polities, some mismatches may remain despite those checks. If you spot one, please report it.
• War and city overlays. Marked wars come from Wikidata's participant records, drawn only where both sides can be matched to polities on this map (unmatched wars still appear as event markers); very long-standing conflicts and sub-battles of a war already shown are folded away to keep the layer readable. City dots use Wikidata population series with point-in-time qualifiers — sparse before 1800, interpolated between points, and used only to scale dots. Do not cite the dot sizes.

How events are chosen, and the bias in that

Events come from Wikidata categories (battles, treaties, foundings, plagues, …), and the panel surfaces the ones near your chosen year. They are ranked by sitelink count: how many Wikipedia language editions cover the topic. Each polity's selection is its milestones plus its most-covered events, distributed across its history to avoid leaving entire centuries unrepresented.

For about thirty countries, the selection is also checked against English Wikipedia's human-curated history and timeline pages to catch omissions. That comparison does not change the ranking.

Sitelink count measures popularity, not historical importance. Wikipedia's editor demographics shape what gets covered, so this map has more events for European and anglophone subjects. The ranking is not neutral.

Notable people are ranked the same way, with one deliberate thumb on the scale. A polity's people come from Wikidata citizenship and ethnic-group claims, ranked by that same sitelink count. Left alone, that produced panels where a country's ten most notable people were ten living celebrities — “association football player” is the single most common occupation among the 75,000 people in our data, ahead of “politician”. So the ranking now lets at most two people whose recorded occupations are entirely athletic keep a place near the front of a country's roster; the rest move behind the writers, scientists and statesmen, which on a full roster usually means out of the ten the panel shows. The most famous stay (Pelé is still on Brazil's panel, Usain Bolt still on Jamaica's), though the panel lists its ten oldest-born first, so don't expect them at the top of it. The panel's ten are then drawn to spread, not just to rank: at most a few from any one calling, at most a few from any one era, each row carrying a small occupation tag; the full roster, every person in birth order with filters by calling, sits one tap behind the “See all” button. We did not manage to do the same for entertainers: Wikidata's occupation field is a catch-all, so a pop singer is often also tagged “composer” and a footballer “politician”, and no rule we tested could tell them apart without demoting people who plainly belong. Expect musicians and actors to be over-represented on modern countries.

A citizenship claim does not always mean someone lived in a country. Wikidata's claims include honorary citizenship, citizenship through ancestry, and citizenship-by-investment, and because the ranking measures worldwide prominence, these claims sometimes put celebrities who never lived there at the top of a small country's panel. Where the connection can be checked, people whose only tie is citizenship now rank behind those who were born, lived, died, or held office there. On a full panel of ten, they usually drop out of the selection; on thinner rosters they still appear, lower down.

This check applies only to modern countries and people born after about 1930. Wikidata's place records use today's borders, which can misclassify earlier lives: someone born in Dublin under British rule, in Danish Scania, or in a village that no longer exists. Those earlier rankings are left unchanged, and a handful of others are held in place by hand where the rule gets their connection wrong.

Built with AI

This Atlas is a human–AI collaboration. The code, the visual design, and most of this prose were written by Claude (Anthropic's AI) across many working sessions. The direction and the final word belonged to a person: Michael Manning chose what to build, tested each round on real devices, pushed back on what felt wrong, and approved every source, fact, and design decision you see. The curated judgments described above — the roster, the borders, which data to trust — went through that human review regardless of who typed them.

What you're hearing

Every sound in the Atlas is synthesized in your browser with the Web Audio API — no recordings, no samples, nothing licensed. The ambient score is generative and meterless.

Selecting some polities re-voices the score with a family-level cultural nod — a scale, an instrument character, a playing gesture: zither-style bends for the Chinese court world, an Andean hocket, a harmonic whistle over a steppe drone, a unison choir for the Byzantine chant world, among a dozen others. Each voice comes from a research pass on the real instruments, plus a historian check on whether the tradition is attested for those polities at all; a “no” leaves the default score in place. These are not reconstructions of period music and make no claim of authenticity: the voices are synthesized gestures toward each tradition's best-known (often much later) instrument forms (the drone cycle under our Indian-world voice descends from a ~16th-century instrument, voicing far older polities), and the Byzantine-world voice is an abstract ensemble (a wash of voices, never a synthesized soloist) with no ison drone: the drone modern listeners expect under Orthodox chant is first attested in 1584, after the empire fell. Where a polity's musical past can't support even a family-level nod — ancient Egypt, drum-ensemble court traditions a meterless score can't voice honestly, traditions synthesis would caricature — it deliberately keeps the default score rather than receive an invented “ancient-sounding” skin.

Sources & licenses

Data

Code & type

How you can use this

You're welcome to explore the map, share a link, and use it in noncommercial lessons, lectures, videos and study guides. You can also read the code and adapt it for personal or hobby projects.

The code, the original creative content and the third-party material have different licenses, listed below. These are the main rules for using the code:

• If you share a copy of the app or host it for others, keep the credits and license notices, identify any changes, and provide matching source code and build instructions. Any code changes you share or host must use the same software license. You don't have to publish private experiments.
• Your version can have its own name. Keep credit to The World Then, with a link back, and don't claim the original work as yours or imply my endorsement. The software license allows credited personal portfolios, including when applying for jobs or freelance work.
• Commercial uses of the code need separate permission. That includes selling it, charging for access, incorporating it into a commercial product or service, supplying it as paid client work, or earning revenue from ads, sponsorships or affiliate links.
• You can accept voluntary contributions toward documented direct hosting costs, up to the amount of those costs, without offering preferential access, features, advertising or services in return. Other donations need permission under the software license.

One request beyond the license: please don't present the map as a definitive historical reference. Its gaps and approximations are described above.

The original creative material (the curation, design, prose, and sound) is licensed under CC BY-NC-SA 4.0 (attribution, noncommercial, share-alike).

The original source code (the page's markup, styles, scripts, and build script) is licensed under The World Then Noncommercial ShareAlike License 1.0, a custom license written for this project rather than an open-source license. The component-by-component terms are in LICENSE.md. This build is release 217; its complete source, with build instructions, is at github.com/mikeroni/theworldthen. Before release 213, published on October 16, 2026, the code was offered under the PolyForm Noncommercial License 1.0.0. Previously granted permissions remain in effect.

The underlying data keeps its own licenses, listed in the tables above. The border geometry itself stays GPL-3.0 (it derives from historical-basemaps, and that license travels with it), and Wikipedia text remains CC BY-SA wherever it travels. Want to use the Atlas commercially or in ways these licenses don't cover? Ask; the answer to good-faith educational ideas is usually yes.

What this site knows about you

The site has no accounts, analytics, tracking pixels, or cookies. Your browser stores seven pieces of local data: sound on or off, which layers are showing, the color-blind palette setting, whether the single-key shortcuts are switched off, whether you've seen the welcome card, the dice's recent rolls so they don't repeat, and whether you've woken the sky's oldest clock. These stay on your device. Clear this site's data in your browser to remove them.

The map, the timeline, and the included historical data run on your device. Your browser also contacts Wikipedia for article text and Wikimedia Commons for images: opening a polity's panel, hovering an event, or opening the reader can fetch text or images that aren't already included, and an open panel can fetch a new ruler's portrait as the years advance. These requests go directly to those services, under the Wikimedia privacy policy.

The three typefaces and every code library are served from this site itself, without a font or code CDN. The downloadable single-file version is the exception: it fetches its typefaces from Google Fonts because it carries no font files of its own.

The site's server keeps standard access logs. The support link goes to Ko-fi, a separate site with its own terms. There is no separate terms-of-service document for the Atlas; its usage licenses are listed above.

Accessibility

The Atlas is designed to be explored without a pointer: the legend list gives a full keyboard path to every polity on screen, media-style keys drive the timeline, keyboard focus stays visible, and map changes are announced to screen readers through a live region. If your system requests reduced motion, the Atlas disables camera flights, autoplay, and decorative animation.

Keyboard shortcuts

These work anywhere in the app except while an article summary is open; while reading, Esc closes the card. If the single-key shortcuts clash with your keyboard or speech software, the keys button beside the playback controls switches them off. Arrows, Home and Esc always work, and every function has an on-screen control. To go straight to a date, click or tap the big year — or press G — and type one: 1066, 500 BCE, -500 or AD 800 all work, and the BCE/CE buttons cover a phone's number pad.

Color-blind support: the legend's “Color-blind palette” switch recolors the religion and language views with alternate palettes tested against simulated deuteranopia and protanopia, so families that share the screen stay distinguishable. The legend and panel always name each family in words, never by color alone. Polity colors carry no meaning by design (the map key is the label, not the hue), so they need no safe variant. The switch is opt-in: the default look stays exactly as designed. Accessibility audit passes ran in July and August 2026; remaining refinements are tracked for future rounds.

Contact & more work

The Atlas is built by Michael Manning. Questions, corrections, and collaboration are welcome. If the map got something wrong, telling me is the fastest way to get it fixed.