Build one yourself,
one article at a time.

Open the editor →

Why most articles need nothing

Wikidata records where things are and when they happened. Battle of Waterloo has both — 18 June 1815, and a point in Belgium — so adding it costs you one paste and nothing else. Congress of Vienna, the same. What breaks the pattern are the things that moved: French invasion of Russia has a date and no coordinates, because a campaign across a continent is not anywhere in particular. That one asks you.

Erask deliberately does not guess. It could follow the article to the country it belongs to and use that — but a French invasion of Russia would land on the centroid of the Russian Empire, in the Urals. A plausible dot in the wrong place is the worst way to be wrong, so the editor asks instead.

One article brings the next twenty

Every row has a ◎ next to it. Press it and Erask asks Wikipedia what that article names — not what resembles it, what it actually links to — and shows the list with a box beside each one. Tick the ones that belong and they all come in at once.

It is the fastest way to make a story long, because it starts from inside the subject. The articles linked from Battle of Cannae are the Second Punic War: Trebia, Trasimene, Herdonia, Zama. Fifty-six of them, and every single one already carries coordinates — only articles with a place on Wikidata come back at all. The dates for everything you tick are fetched together, in one request per fifty.

They arrive grouped by the section of the article the link came from — Aftermath, Historical significance — which is what makes a long list choosable: the heading tells you why that link is there. Anything already in your story is left out, since offering you things you have decided on is noise.

One thing it deliberately does not do: take a date from an article that only has a founding year. Tarentum links from Cannae and Wikidata dates it 706 BC — the founding of the town, not a moment in the war. That would put a Punic-war story seven centuries early, so the row comes in with the date blank and asks you. A field to fill is better than a wrong date already written.

The order is yours

Drag a row by the handle on its left and the others move aside to show where it will land. The order is the story — it is the sequence the points light up in — and it is not always the order things happened: sometimes you want to open at the end. The button that sorts by date is still there for when you do want chronology; dragging is how you say you don't.

You do not type coordinates

Asking somebody for a longitude is asking them to leave and come back with homework. So the field that wants a place is not a field: press it and the sheet gets out of the way, the globe is yours, and you put the point where the thing happened. Drag it until it looks right. That is the whole interaction, and it needs no idea of what a coordinate is — you already know where Moscow is.

Everything you add shows up on the globe as you go, so the story takes shape while you build it rather than at the end. Dates are typed, in whatever form you have one: a year alone is enough for something you only know the decade of, and a full date when you have it. There is a button that puts the list back in the order things happened, because articles come to mind in an order that is almost never the one they occurred in — and that order is what the globe plays back.

The story is the link

Nothing is stored and there is no account. What you build lives entirely in its own address:

That block is the whole story, packed. A Wikidata identifier is a number, not eight characters of text; a story is a path, so each place is written as the step from the one before; the dates are in order, so they grow by a little at a time. Seven events fit in 135 characters, fifty in about 750 — roughly a third of what the same story used to cost.

It is no longer readable at a glance, which was a real thing to give up. What it buys is length: a story can now run to several hundred events before the address stops fitting where addresses have to fit. Links written in the older, readable form still open, and always will — there is nowhere else a story exists.

The same thing an AI does

The prompt for language models produces exactly this format. So a story you build by hand and one written by ChatGPT are indistinguishable when opened — and you can take a generated story, press Edit, and fix what the model got wrong. That last part is the reason the editor exists: models are fast and confidently inaccurate, and until now there was no way to correct one.

What it will not do

• Up to three hundred events. Past that the timeline becomes a comb, and resolving the articles stops being a couple of requests.
• Ten captions, no more. They are the one thing in a link that is real text, so they cost far more than an event does — ten of them still leave room for hundreds of events, while one on every event would stop you at under two hundred. Ten lines you chose say more than a hundred written out of duty.
• Every event needs a date and a place. The editor will not let you copy a link until they all have both — the globe would silently drop the rest, which is worse than being stopped.
• Curated stories cannot be edited. They are checked by hand and live in files; opening one here would make a disconnected copy, not a correction.
• Nothing is saved. Close the tab before copying the link and the story is gone. Copy early.

Where to start

Pick something you already know well — a war, a life, a band's tours — and add five or six articles. The interesting part comes when you press play and see them in an order you knew but had never watched.

Open the editor →