Atlas causal

How the atlas is drawn, in full. The about page has the short version.

about narratives sources

What this is

The atlas is a map, a timeline and a graph of the same records. Events are the points of the graph. An edge is a link from one event to another. Each edge has a type, a confidence level and a written argument for the link. You pick an event and follow its consequences forward. The atlas then shows which other branches also led to the event you reached. This is what the format is for. It keeps the path you happened to walk from looking like the whole explanation.

Time is a window of years. The shaded band on the timeline is the window. Drag either end to widen or narrow it, drag the middle to slide it, or double-click a year to snap it to that decade. The arrow keys move whichever end has the focus. The map shows every event whose own years overlap the window, so an event that lasted ten years appears in every window it touches. The map draws the territories of the window's last year. The / key jumps to the search box. Search finds events by title, actors by any of their names and places by any of theirs.

An event happens at a place, and a place is a record of its own. Everything that happened in Lisbon points at one Lisbon record, so one correction fixes all of those events. Click a place name, on an event or at the head of a stack of marks on the map, to open what happened there, in order. Events outside the window stay in that list, faded. The place also lists the actors who appear there most.

Actors are the people, states and institutions that take part in events. Actors also have relations with one another. An edge cannot express these, because an edge joins two events. A relation has dates and a type: a regime of a state, a body part of another, a person or a state that is a member of one, a state succeeded by another, or two actors allied. It cites its sources like any other argument. Relations have no page of their own. The card of the actor at either end lists them. Portugal's card names its four regimes, and the Estado Novo's card says which state it was a regime of and what belonged to it.

A post someone held is an office, with a tenure for each turn at it. The atlas does not use a relation here, because one person may hold one post more than once. An office has its own card, and the search box finds it. The card names the actor the post belongs to, the kind of post, and every turn at it in order. Each row opens the person who held it. On the card of an actor that owns posts, each office is also a strip of its holders across that actor's years. A century of prime ministers then reads as a row of bars, and clicking a bar opens the person. An office states only that the post existed, so it cites nothing, and its card says so. The tenures carry the sources.

A narrative is one person's account of the graph, read step by step. Each step points at an event or a link and carries the narrator's paragraph on why that step follows. The map, the graph and the timeline move to each step, and the whole walk stays marked so you can see where it goes. A narrative changes nothing in the records it walks through. Its author signs it and cites sources like everyone else. Where two accounts of the same period disagree, both stay and you can compare them. The list of narratives is in the header beside the bibliography. It groups them by the centuries they cross, so two accounts of the same years sit side by side. Every event also lists the narratives that pass through it.

The first records covered Portuguese expansion, because its consequences cross continents. The atlas now runs 1415 to 2026 over five regions, and it keeps growing. If the structure of the records works for one period, it should work for others. The way to find out is to try it on another period.

The hard part was never the map. No dataset states that one event caused another across four centuries and two oceans. A person has to write that, with sources, and another person has to read it and put their name to it. So this is editorial work with a map attached. One person makes it, with an assistant. Most records start as drafts made from Wikipedia and Wikidata, and each card credits the article and the revision it came from. Every causal link carries a written argument and at least one source. Adding ?review=1 to any address shows how far a person has read each record. The reading is never handed over. A model will write a fluent causal story about any pair of events, even unrelated ones. On a history site that is total failure. So a link is only as good as the person who signed it.

Every claim rests on a cited source. A source is a record too. The atlas stores it once, and every record that cites it points to it. Click the title in any citation to open that work and everything in the atlas that rests on it, with the page each record names. A work cited in dissent, against a link, appears apart from the works cited for it. The bibliography lists every source. The build makes it from the records, and nobody edits it by hand.

How to read it

How to read a card

Clicking a mark on the map, a bar on the timeline or a point in the graph opens a card in the panel on the right. Every card has the same parts, for an event, an actor or a place.

At the top is the head. It holds the title and one line saying when and where the event happened and which lane it is in. A lane is one row of the timeline. On an event the head also shows the actors as small chips. Hover a chip, or reach it with the keyboard, to see what that actor did in this event. Click it to open the actor. Under the head are three small links: the full entry, the Wikipedia article, and a way to disagree with the record.

Then comes the summary, the atlas's own paragraph about the record.

Everything else sits under headers you can open, each with a count beside it: Consequences, Causes, Other branches, Sources, Part of, and on an actor Relations and Territory. The count tells you what each header holds before you open it, and one click opens it. Only one is open at a time, and the atlas remembers which one you chose last. If a section holds a disputed link, its count says so (“Consequences (3, 1 disputed)”), so a closed section never hides an argument.

Which section opens by itself depends on what you just did. Walking a chain opens Consequences. Arriving from a source’s card opens Sources. Otherwise the atlas opens the one you had open last.

While you follow a chain, the path you have walked appears as a breadcrumb above the card. Each earlier step links back to itself. Going back to a step in the middle drops the steps after it. The path is the argument you are following, and you have not taken those later steps.

Some events are conditions that last, such as “Angolan economy dependent on oil, 1975–”. A condition is a process with no single day and often no end. The atlas needs them because a question like why is Angola poor? has to end at something, and a battle is the wrong answer. People write conditions like every other record, with the same sources and the same links. Sometimes the atlas puts a path together itself, without you clicking each step. The card then says so in a line above the record. Every step of such a path is still a link a person wrote, with its confidence and its dispute marks unchanged.

Opening a record (an event, an actor, a place, a source or a narrative) adds an entry to your browser’s history, so Back returns to what you were reading. Moving the view adds no entry. Panning, zooming, dragging the time band or changing the lanes replaces the current entry. Otherwise Back would step you frame by frame through a drag. The card names its own Back link, so you can see where it goes before you press it.

The five edge types

An edge is one of caused, enabled, reacted-to, precondition-of or inspired. There is no general type, and there never will be. Folding the five into plain cause and effect would make history look simpler and more inevitable than it was. A reviewer checks that each edge has the right one of the five.

The three confidence levels

The atlas defines confidence by evidence, so anyone can check it.

consensus
Historians accept the link. It needs at least two cited sources by different authors. Two sources are the minimum. A reviewer still judges whether they really agree.
probable
The cited sources support the link, and no known work argues against it. Most edges have this level.
disputed
Qualified historians disagree about the link itself: whether it exists, what type it is, or how much weight it has.

What did this lead to by a given year?

The consequences list on an event goes one step out: what this event led to directly. Under it, what did this lead to by a year asks a wider question. It lists everything downstream, however far, up to a year you choose. The list is in order of the number of steps to each event, then of date. For each event it gives the type and confidence of the first step. It marks any path that runs through a disputed link, so a long chain of weak links never looks like a short chain of strong ones.

The year starts at the end of the window, and you can set it to any year. Choosing one of the answers walks the shortest path to it, and that path turns red. You can then ask what else led to the event you reached. That is the question about other branches again, asked from the other end. While you have a year of your own set, the map, the graph and the timeline light up everything reachable, fainter with distance.

A disputed edge

A disputed edge stays in the atlas. Removing it would mean taking a side, and the disagreement itself is worth seeing. The map draws it as a dashed line and the panel marks it. Following a chain across one always tells you so, and shows who disagrees and what they cite. The atlas keeps the citations against a link apart from those for it, so a book that argues against a link never appears as evidence for it.

Nobody deletes a record. A record that turns out wrong or duplicated becomes a tombstone, which points at the record that replaced it. Old links still lead somewhere, so a URL someone shared keeps working.

An event inside another event

An event can be part of a larger one, such as a battle inside a war or a decree inside a revolution. The smaller event's card says what it is part of. The larger event's card lists its parts in order. Each links to the other. Being part of something says only what an event was inside. It says nothing about causes. A cause is an edge, with an argument and sources.

An event with parts has a second, thinner ring around its mark. It has the ring on the map, the timeline and the graph, at every zoom and in every grouping. The ring means there is more inside, so open the event to see its parts. Focus on this on the larger event's card then keeps that event and everything inside it.

Where there is room, two views do more. The timeline draws a larger event whose parts share a lane as a thin rule along the top of that lane, across them. The graph uses it as a level of detail. Zoomed out, the graph draws an event's parts inside it, as one mark with the count of what it holds, and links to a part run to the larger event instead. Zooming in separates them again. The graph never folds away what you are using: the open event, the path you have walked, a lens or a narrative's steps. No mark moves when the others fold, so the picture does not rearrange itself under you.

Large events drawn as an area

Some events are too large for one mark: a world war, a pandemic, a long regional crisis. The atlas draws such an event as a shaded area when a contributor has written how far it reached, or when its parts fall in more than one lane. On the timeline it is a pale band the full height of the drawing, under the bars, named on the axis. On the map it is a wash over its lane. An event that spans the whole map gets no wash, because a tint over everything would cloud every coastline and territory. The map names it in its corner instead. The bar and the mark stay where they were, so you can still open a large event and reach it with the keyboard. Its card says which of the three cases it is and why.

Reach describes the event and does not make its mark bigger. Mark size still comes from the number of links and actors the record holds.

In the map's bottom-left corner is the count of events in this window that have no place at all. These appear on the timeline only. The count tells you what the map leaves out.

The lens

The card of an actor, a place or a source offers show only these. This puts a lens on the whole atlas. A lens is a filter for one record. The map, the graph and the timeline then show only the events that record is involved in, with the links between them. Selecting a record is different. Selecting highlights it and keeps everything else in view. A lens removes the rest. The header says which lens is on and offers the way out. The address bar carries the lens too (?focus=actor:salazar), so you can send it in a link. The window of time, the path you have walked and the question about a year all still work inside a lens. The lens does not narrow search. Search still finds and lists a record the lens leaves out, and marks it as outside.

Where the map is centred

The map opens on the whole world with 150° east in the middle. Asia and the Pacific are at the centre, and the Atlantic falls at the two edges. This is a deliberate choice. The atlas is a map of human history, and Europe is only one part of it. Every projection has to put something at its edge. This one puts 30° west there, the meridian half a world from the centre, and cuts the map along it. A shape lying across that line would become a smear from one side of the picture to the other, so the import cuts every outline there.

The atlas measured where to cut. Of the centres from 140° to 170° east, 150° east is the only one whose cut line, 30° west, crosses no land except Greenland and Antarctica, which no choice avoids. The line runs down the middle of the Atlantic, between Flores and the rest of the Azores, four and a half degrees east of the corner of Brazil and twelve degrees west of Africa. The table of candidates is in STATUS.md. The projection is one constant in one file. The map uses no tile server, no map library and no API key. It draws its own coastlines from a public-domain file, in a projection written for this site.

The map and the timeline together

Pan or zoom the map and the timeline follows it. Its lanes then hold only the events whose place is on screen, and a small line above them says how many of the total are in view. The timeline always keeps the event you have open and the path you have walked, even when the chain runs off the edge of the map. Show the world puts every event back without moving the map. The address bar holds the visible area (?bbox=-10,36,-6,43), so you can send a view of one coast in a link and it opens on that coast. The area is always in real longitudes, and it may run east past the 180th meridian and out the other side. ?bbox=170,-20,-170,0 is Fiji. With the map centred on Asia, that is an ordinary view near the middle of the picture. The graph has no visible area of its own, so the map never narrows it, and the graph says so while the map holds one.

The mouse wheel over the timeline narrows or widens the window of time around the year under the pointer. Dragging the timeline's empty ground slides the window. The lanes never move, because they always cover the whole span of the data. A click on empty space, such as the sea or the empty ground of a lane, clears what you were holding: the open event and the walked path. Dragging never clears them, and a click on a territory still opens its actor. You can drag the edges between the map, the timeline and the panel to resize them. A double-click on an edge restores it. Your browser keeps those sizes. They are not in the link, because a link records what you were looking at and leaves out how you arranged your window.

What the lanes are

The timeline's lanes and the graph's bands are the same thing: a grouping of events into rows. You choose the grouping next to the Map and Graph buttons in the header. The default is no grouping. The timeline then packs the bars into as many unlabelled rows as it needs for none to overlap, and the graph has no bands. The other three groupings give one lane per actor, per place or per region. For actors and places the lanes are the six with the most events in the window, plus Other. The picker lets you choose your own lanes and put them in your own order. The link carries the choice (?group=actor&lanes=salazar,paigc).

The atlas draws each event in exactly one lane. An event with four actors is one event. Drawing it four times would turn a count of events into a count of participations without saying so. A fixed rule picks the lane: of the event's actors that have a lane on screen, the one with the most events in the window. The event's card always says which lane it is in, by what rule, and what else it involves, so you can check the rule.

The graph view

The Graph button in the header replaces the map with a third drawing of the same records. Every event is a point and every link a line, so you can see the whole web at once. It is the same atlas, with the same panel, search, window of time and link in the address bar.

Left to right is the year, on the same scale as the timeline, so you can see time move in one direction. No simulated physics places anything. Down the page runs one faint band per lane, following the grouping, and there are no bands without one. Inside a band, the graph spreads out the events of each year and orders them so the lines between them are short and cross as little as possible. The window of time is the shaded band across the picture. Events outside it fade and stay on screen, so the whole web is always there. Drag to move, scroll to zoom, double-click to go back. Zoomed in, every point on screen shows its name.

Zoomed out, the graph draws events too close to tell apart as one mark, with a +3 beside it saying how many more are underneath. The links between two such marks become one line. The line is heavier the more links it carries. It takes the commonest of their types and the ink of the least sure of them. The graph drops nothing, and the counts add up to every event. Zooming in splits the marks. Clicking one lists what it holds in the panel and zooms to where it comes apart. Events in different bands never merge. Neither does anything you are working with: the open event, the path you have walked, the branches drawn at it, a lens, a horizon (the year you set for what an event led to), a narrative's steps and the events of an open actor all keep their own mark.

Clicking a point opens it in the panel. Clicking a point that the open event leads to walks the chain, the same as the panel's follow button. The graph, the map and the timeline all draw the path you have walked in red. When an event is open, the graph fills in the other branches that led to it, the ones you did not walk. This is the question about other branches, drawn as a picture.

The key to the lines

The five types differ in the pattern and weight of the line. They share one colour, because the atlas keeps its one accent colour for the path you are following. The ink of a line shows how sure the atlas is of the link: solid where historians agree, fainter where the link is probable, faintest where it is disputed. Type and confidence are two separate features of one line. A disputed caused link is still an unbroken line, only faint. The card says all of this in words too. The key lets you see it without opening a card.

caused
enabled
reacted-to
precondition-of
inspired
how sure: consensus, probable, disputed

Territories

The territories layer shows who held which ground in one year: the last year of the window, which the band on the timeline shows. It covers 1400 to 2019 from two sources, with a visible seam at 1886 that this page explains. After 2019 it shows the last year the sources cover and says so. Before 1400 the layer is empty, and the atlas does not guess to fill it. The switch in the header turns the layer off. The coastlines have no switch. You read everything else against them, and without them the map is a scatter of dots.

The outlines from 1886 on come from CShapes 2.0: Guy Schvitz, Luc Girardin, Seraina Rüegger, Nils B. Weidmann, Lars-Erik Cederman and Kristian Skrede Gleditsch, Mapping the International System, 1886-2019, Journal of Conflict Resolution 66(1), 2022, doi:10.1177/00220027211013563. The atlas uses them under CC BY-NC-SA 4.0, which differs from the licence of the rest of the data. The atlas simplified the coordinates to make the files smaller. Every date, status, sovereign and capital is the dataset's own.

Territories before 1886

The outlines from 1400 to 1885 come from Historical Basemaps, by André Ourednik and contributors. The atlas uses them under GPL-3.0. That licence differs from the rest of the data and from the outlines after 1886. The dataset is a set of snapshots. It draws the whole world at 1400, then at 1492, 1500, 1530, 1600, 1650, 1700, 1715, 1783, 1800, 1815, 1878 and 1880, and says nothing about the years between. Its authors call it work in progress and ask users to check it against other sources.

So the atlas draws every outline before 1886 with a dashed line all the way round, coast included. Nothing else on this map draws a line along a shore. The dashes mean the whole line is uncertain. Move the window to 1625 and you see the 1600 snapshot, because that is the last border the dataset drew. Move it to 1750 and you see 1715. The atlas marks these outlines probable, the word an edge carries when its sources support it and no known work dissents. It marks the territories after 1886 consensus. A solid outline is a border somebody recorded. A dashed one is a border somebody drew once, which this map keeps until the next drawing.

This source does not say two things, so the atlas does not ask it to. It has one field for who held authority over a region, and that field mixes a colonial power, an overlord and the region's own name. So no territory before 1886 appears as anybody's dependency. Those centuries have no tinted possessions and no dotted borders, and an empire and its colonies are separate washes. The source also lists kingdoms, confederations and peoples under one heading without telling them apart, so the atlas calls nothing here a state. Both are gaps in what the map shows. A gap is better than a wrong answer that looks confident.

The snapshots are also far apart. The gaps reach ninety-two years (1400 to 1492), seventy (1530 to 1600), sixty-eight (1715 to 1783) and sixty-three (1815 to 1878). No openly licensed source for these centuries has more.

What the territory styles mean

A state holding its own ground is a solid wash with a thin border in a darker shade of the same colour. A dependency (a colony, protectorate, mandate or occupied territory) takes the lighter tint of whoever held it, with a dotted border. An empire and its possessions then read as one family, and hovering names both. A dashed border means the presence itself is disputed, the same as a dashed link. A wash dashed all the way round, shore included, means the outline is probable: a snapshot, as the section above explains.

The map draws borders inland only. CShapes has its own coastline, which differs from this map's. A territory outlined all the way round put a second shore beside the first, a few tenths of a degree out and plainly visible when zoomed in. The import keeps every boundary once, as a line two territories share, and each period's file carries that list beside the outlines. The map strokes only the boundaries a territory shares with a neighbour. The shore is the Natural Earth coastline underneath and nothing else. So an island state such as Japan, Iceland or Cuba is a wash with no edge of its own, and Switzerland has a border all the way round because it is inland all the way round. One result is worth stating. Where the only thing across a border is a state that Gleditsch and Ward do not list, the source does not say who is there, and the map draws no line. Switzerland's boundary with Liechtenstein is the case.

What the colours mean, and what they do not

There are eight territory colours and no legend for them. The layer control is the legend for everything else. A colour says only one thing: this is not the same territory as the one beside it. It does not say which polity, what kind of state, whose empire, or how large. Two hundred polities share eight colours, so a colour legend would mislead. To see who was where, hover a territory or click it. Clicking opens the actor's card and fills in everything that actor held.

The records do not store colours, and no person picks them. tools/build-palette.mjs works them out from the borders. It finds who touches whom in every period the outlines cover, then colours them so that no two neighbours ever match. It writes the result to data/geo/palette.json. The colour belongs to the actor, so a territory keeps its colour as the years pass. Angola has the same colour in 1975 as in 2019. Before 1975 it has Portugal's colour, lightened, as a dependency of Portugal. Sometimes eight colours are not enough: an empire that borders a quarter of the world cannot differ from every neighbour. The file then records which neighbour shares its colour.

The territory colours sit under two more important ones. The atlas fills the actor you have selected in cobalt over every territory colour. It draws the path you have walked in madder red over that.

Category symbols on the marks

An event may have one of twelve categories: war, treaty, election, revolution, law, founding, disaster, economy, culture, science, death, other. When it has one, a small line symbol sits over its mark on the map and at the left of its bar on the timeline. The mark itself does not change. It is still the circle you click, and the symbol takes no click of its own. An event with no category keeps the plain circle, and most events have none. Someone filled in categories for part of the records, and nobody has done the rest yet.

The layer control is the legend. Under “events by category” in the options it has one row for each category that has records, each with its symbol. That is where you learn the shapes. The map has no second legend, for the same reason it has none for colour: a panel of twelve shapes would cover the picture.

Turning a category off narrows the map, the timeline, the graph and the count in the corner together. A category switch only affects events that have a category. Turning “war” off hides the wars. It does not hide events with no category, which stay until you switch off the whole “events” layer. The link records what you switched, so you can send a narrowed view.

The base map

Under the borders and the marks is a base map: the coastline in detail, rivers, lakes, physical regions such as deserts, plateaux and ranges, peaks, and cities. It comes from Natural Earth at two levels of detail. The coarse level covers the whole world. The fine level covers the grid square you are looking at, so panning across the Atlantic never loads Asia. Each feature also carries the zoom at which it is worth showing, so a river appears as you zoom in. The base map is the quietest layer and takes no clicks. A click on a river counts as a click on the sea and clears what you were holding.

One rule places every name on the map. An event's name wins over a city's, and a city's over a river's. The map leaves off a name that does not fit and never nudges it aside, because a moved name would end up pointing at its neighbour. A name appears after its dot and never before it, so as you zoom in the shape comes first and the word after. Rivers, lakes, ranges and peaks use the same type size as everything else, with slightly wider letter spacing. The map has only one type size.

A city may show the name it had in the window's year. A place record may say what a place was called and when, for example Lourenço Marques until 1976 and Maputo after. The map then writes the name for the far end of the window, and moving the window changes the name. Hovering shows all the names, with the years of each. The atlas invents none of these names. A former name is a claim about the past, so it appears only when a person has written it into the record, with the years they meant. Where no record gives one, the map writes the modern name and says nothing about the year. Today no place in this atlas has a dated name, so nothing on the map is dated yet. The code is ready for the first one someone writes.

Some places the atlas names are missing from world gazetteers, such as a parish of Lisbon, a district of Mozambique or a field near Villanueva del Fresno. The map writes these from the atlas's own record, at the point the record gives. They take priority over large cities the map knows only from Natural Earth, because this is a map of what happened in these places.

One control switches everything the atlas draws, and that control is the legend. It has four parts: territories, events, a group called base map with one row per layer (rivers, lakes, physical, mountains, cities), and a group called events by category with one row per category and its symbol. Each base map row has a small swatch in exactly the colour and line the map uses. Both groups start folded, so on a phone the drawer has four things to tap instead of nineteen. The swatches and the symbols are all the map explains about itself. There is no second legend over the picture, for the same reason there is none for colour.

The coastlines have no row, under either of the two names the code gives them. You read everything else against them. The detailed shore is the same line at a finer level, so a switch for it would only switch a level of detail. A layer you switch off costs nothing: the map does not draw it and never fetches its data. The link records what you switched. A link that names a shorter list means these and nothing else, so ?layers=territories,events opens with the base map off.

A border is still a modern idea applied to the past. The map draws borders only for a period that had them and only from a source that dates every change. Earlier periods will need loose zones instead of outlines, and the presenceType field exists for that. The timeline's lanes by region are continents. The atlas works them out from the coordinates of each event's place and does not store them on the record, so redrawing the lanes never means rewriting records.

The full entry

A card shows a summary: a sentence or two, enough to know what the record is while you follow links. Under it, on every event, actor and place, is Read the full entry. This opens a page of its own at entry.html?id=…. The page has the record's dates and place, its actors, its relations, the narratives that pass through it, everything it rests on, and a long text about it.

A person writes that text from sources they cite. It does not send you to another encyclopedia. Most records have no entry yet, and their pages say so and ask for one. The atlas is a graph first, and the long texts arrive one record at a time as people write them.

An entry uses a small part of Markdown: paragraphs, two levels of heading, emphasis, lists, quotations, links to other records and to the web, and citation marks like [^some-source p. 12]. Each mark must name a work the record already cites. Entries allow no raw HTML and no images. Everything under data/ is someone's contribution, and letting a contribution write markup would let it put anything on the page. The page numbers the marks in the order the entry first names each work, and each mark links down to that work in the entry's list of sources.

Disagreeing with a record, and exporting a picture

Every card ends with Discuss this record. It opens a correction issue on the repository, named for the record and carrying the address you were looking at, so whoever answers sees the same picture. A record is an argument someone made, and you can disagree with it in public.

The map and the graph offer Export as SVG. This saves the drawing on screen as an SVG file, with the stylesheet and the colours your browser computed written into it, so it opens anywhere. The file is only a picture. The data is the repository, in full, under the licences below.

On a phone

On a narrow screen the atlas stacks its parts. The map or the graph fills the screen, and the timeline is a strip under it. The panel becomes a sheet that comes up from the bottom edge when you tap a mark. When it is in the way, drag or tap its handle to push it back down. The search takes a line of its own, and the grouping and the layer switches are behind Options. The map and the graph both pan with one finger.

A link opened on a phone shows the same records as on a desktop. Pane widths, which section of a card is open, and whether the sheet is up are yours and stay out of the link.

The link to Wikipedia

An event, an actor or a place may carry the identifier of the matching Wikidata item and the titles of the Wikipedia articles about it. Where there is a title, the card offers Read more on Wikipedia. It links to your language when an article exists in it, otherwise to English, and it always says which. The search box also finds the record by the article's title.

That link leads out of the atlas. Everything the atlas states is in the record: the summary, the argument on each link, the confidence and the sources. Nobody writes these from an encyclopedia or checks them against one. No record holds an article's text, only its title, which is a name. The identifier lets a machine find the record again and lets a reader go and read another account. The atlas never counts it as evidence. An encyclopedia reports scholarship and is not scholarship itself. So a link that cites only Wikipedia cannot be consensus, and the validator refuses it. The number of language editions with an article changes nothing in how the atlas draws anything.

Licences

The atlas uses several licences, because one cannot fairly cover code, prose, two imported datasets and two typefaces.

Why the licences are final

Changing a licence needs the permission of everyone who holds copyright in the work. Once a stranger's paragraph is merged, that includes the stranger, and the group grows with every contribution. Soon nobody could find and ask all of them. So the licence is a decision made once, before contributions opened, and it is final. The same reasoning keeps imported geometry under an incompatible licence in its own directory, with its own attribution, and out of every record a person wrote. CShapes is the case: CC BY-SA cannot absorb its non-commercial clause in either direction.

Contributions

Contributions are not open yet. The project is still testing the schema, the structure every record follows, on the first period. Moving a hundred contributors' records to a new schema is far harder than moving your own. So the project builds and tests the way to contribute first, and announces it after.

When it opens, it will work like this. A form builds a bundle: an event, the edges that connect it, and any source not yet in the bibliography. Before you send anything, your browser checks the bundle against the records already here. It checks that every reference leads to a record and that no link runs backwards in time. The form copies the bundle to your clipboard and opens one issue. A maintainer reads the issue. Only if they accept it does it become a pull request. A person still reviews the argument, the sources and the confidence before merging it. The maintainer merges nothing without sources. GitHub carries the issues and pull requests, and a person makes every decision.

CONTEXT.md in the repository explains the reasoning behind all of this. ARCHITECTURE.md describes the structure, and STATUS.md says what is built so far.