Methodology

What readers see

Each timeline entry is a short attributed excerpt with a date shown at the precision the source actually supports (year, month, or day), an outlet, confidence label, live source link, archive status, publisher trust tier, and corroboration count. We do not publish full articles or private page snapshots.

Capture pipeline

1. Poll allowlisted outlets (RSS, official pages, permitted APIs).
2. Fetch the page without bypassing paywalls or logins.
3. Submit the URL to the Internet Archive (Wayback) for a public archival copy.
4. Store a private PDF/screenshot vault copy for integrity checks only.
5. Extract dated statements with attribution confidence scores.
6. Auto-publish when source trust and confidence gates pass. Borderline confidence and ambiguous life/legal subject matches are stored as HELD for an internal review queue rather than shown publicly. Unrated rumours are discarded entirely.
7. After you add a figure, a durable collect job is queued (Cloudflare Queue and/or a Node collection worker with a Postgres lease). The job searches Wikipedia, Wikidata, Wikiquote, then follows Wikipedia citations only to an allowlist of trusted outlets (Reuters, AP, BBC, NPR, major papers, .gov / UN / WHO, etc.). Confidence rises with publisher trust and multi-source corroboration. The timeline UI shows Queued until a worker claims the job, then live step progress.
8. The same job parses campaign and election sections for politicians, so candidacies stay distinct from offices held, and parses the article’s “Honours” section for trophies and awards.
9. Photos, audio and video are collected from Wikimedia Commons and shown with their author and licence. Files marked fair use or otherwise non-free are skipped rather than re-published here.
10. Headlines under “In the news” are live links read from publishers’ own feeds, plus Guardian and New York Times archive search with accent-aware name matching. A headline relevance gate requires word-boundary name matches (and context for bare surnames) before a story is attributed to a figure.
11. Sayings are gathered from every Wikiquote edition that has a page for the figure. Sections holding quotes abouta person are skipped so other people’s words are never attributed to them. Undated sayings are marked approximate rather than stamped with today’s date.
12. Specialist databases are chosen from the figure’s roles: TheSportsDB for trophies and transfers, MusicBrainz for releases, Crossref for research, Hansard for what MPs said in Parliament, and the Nobel Prize API.
13. Social posts come only from accounts Wikidata records for that person. We never search a network for a matching name, because an impersonator’s words must never land on someone’s timeline.

Events nobody reported

Wikipedia readership and editing activity are recorded alongside the timeline. When a figure’s article is read many times its normal volume in a single day and editors converge on it, that day is marked as an attention spike — paired with a headline from that week when one exists. It catches moments that no feed we subscribe to covered.

Rumours and contested claims

We do not collect from tabloids or gossip sites. Repeating an unverified allegation about a living person is the clearest legal and ethical risk this project has, and no amount of traffic justifies it.

Personal life, relationships, and legal history come from Wikipedia sections that its biographies-of-living-persons policy already governs. Sentences that only weakly identify the figure — especially in legal or controversy prose that also names other people — are held for review rather than published as factual records. Circulating claims come from the Google Fact Check corpus, which supplies each claim together with a published rating. Those entries are marked Reported claim, always name the publisher and the rating, and are never asserted as fact. A claim with no rating is discarded, not shown.

Link rot

When a live URL fails, the UI prefers the Wayback link. Private snapshots remain offline evidence for operators, not a public mirror of publisher content.

Corrections

Use “Report issue” on any statement. Reports land in an internal moderation queue alongside held extractions. Operators with the admin secret can publish, hold, or retract items. Retracted items leave the public timeline. Authenticated follows and personalized alerts are not part of the public product yet.

Legal note

This product is designed around short quotations, attribution, and source links. It is not legal advice. Publishers retain copyright in their articles; respect robots rules and terms of service.