WhatLedTo

Data

Every timeline as JSON and Markdown

The same entries, sources and quotes that the pages show, in a shape a program or an AI agent can read. Data version 1.

Where the files are

A timeline page also answers content negotiation: request /events/<slug> with Accept: application/json or Accept: text/markdown and you get the JSON or Markdown file, with Content-Location naming it. Every page links its alternates from its <head>. The JSON files allow cross-origin requests from any page, and, unlike the HTML pages, they are not rate limited.

What a timeline file contains

Top level: slug, title, url, vertical and topic, tags, tldr, created_at and updated_at, then the sections below. Timestamps are ISO 8601 in UTC; event dates are YYYY-MM-DD, or YYYY-MM / YYYY when date_precision says the day or month is not known.

nodes
The entries, oldest first. Each has an id that is stable for the life of the entry, a url that addresses it on the page, the date the thing happened, a title, a two-to-three sentence summary, a type (announcement, launch, regulatory, legal, financial, report, response, delay, shipment or other), the entities it is about, published_at, and an editor's note when a date was corrected or a source disagreed. sources lists every page the entry was taken from with its url, outlet, publication date and the verbatim quote that supports the summary.
entities
The parties on this timeline: name, kind (company, person, regulator, product, other) and a one-line role. Names match the strings used in each entry's entities.
relations
Links between two entries, from the earlier to the later, that the dates alone do not show: causes (a source says the earlier brought the later about), responds_to, contradicts (with the disagreement spelled out in note) and follows (the later completes what the earlier began). Every relation carries a note saying what the link is. Treat causes as the editors' reading of the sources, not as a fact of the same standing as an entry.
links
The same three explanatory kinds, but with one end on another timeline. Each end gives its event slug, entry id, date, title and url, so a link can be followed from either file. A link is written once and appears in both timelines' files. The same rule of evidence applies: it exists only where one entry's own text names the other's fact.
key_node_ids
The entries that carry the story, chosen when the summary is refreshed.
catalysts
Dated things to watch next, each with a why and, where one exists, a source_url.
open_questions
What the sources have not settled yet.
metrics
Series charted on the page, such as a price or a rate: key, name, unit, kind (step, line or bar) and points. A point exists only where a published entry states the value verbatim; each carries the node_id and the quote it was read from.
related
Timelines the editor has linked to this one.
changelog
Every change to the timeline, newest first: research passes, published entries (with their node_ids), summary refreshes and corrections. Entries are never silently edited, so a diff of this list is a diff of what the site knew.
hero_image, hero_credit, image, image_credit
Photographs and the attribution they are shown with. They are licensed to this site as described under Images; the credit travels with the file, the licence does not.

Not included, on purpose: unpublished candidate entries, the research brief and search queries each timeline runs on, and the instructions the pipeline uses to read chart values. Those are how the site is made, not what it says.

Stability

Fields are added, never renamed or removed, within a data version; data_version in every file changes only when a field changes meaning. Entry ids do not change. A timeline that is retired keeps its files with status set to archived. Fetch /index.json and compare updated_at to know which timelines to re-read; the RSS feed and each file's changelog say what changed.

Terms

The reuse terms apply to the files exactly as they apply to the pages: quote and cite entries with a link to the entry's url, use the data in research, articles and AI answers with attribution, and do not republish whole timelines or train models on the text. Every response says so with a Content-Signal header. For a licence beyond those terms, or to ask for a field or a format, write to editor@whatledto.com.