The Founders Magazine

Methodology

Sources & methodology

Founders Magazine is an automated desk with human editorial control. This page describes exactly how an item travels from a public source to a published story.

The pipeline

  1. 01Scheduled fetchPublic feeds, official newsrooms, regulator and filing APIs are polled on a schedule. Nothing is dropped at intake.
  2. 02Source normalisationItems are mapped to a common shape: publisher, title, URL, timestamp, primary/secondary flag.
  3. 03Dedupe & event clusteringItems describing the same event are grouped. Clusters keep every member, so source count is real, not assumed.
  4. 04Relevance scoringCandidates are scored for editorial relevance. Low-scoring items are set aside with a logged reason — never deleted.
  5. 05Fact extractionAmounts, investors, names, dates and roles are extracted with a reference back to the source that stated them.
  6. 06Source verificationFacts are checked across cluster members. Anything supported by only one secondary source stays flagged.
  7. 07Article generationAn original summary is drafted from the verified facts. Source phrasing is not reused.
  8. 08Image prompt & generationAn editorial cover image is generated for the story and labeled as an illustration.
  9. 09Editorial QA rulesMinimum sourcing, disclosure, and no-invented-facts checks run. Failures block publication and are shown to the editor.
  10. 10Publish or scheduleOnly drafts that pass QA can auto-publish. Everything else waits for a human.

Sourcing labels

Primary source

Official statement, filing or regulator data. Can support publication on its own.

Multi-source

Two or more independent sources agree on the facts.

Single source

One credible secondary source. Published only with this label attached.

Unverified draft

Reported but unconfirmed. Drafted, never auto-published.

What we never do

  • Invent or estimate funding amounts, valuations, investors, founders or quotations.
  • Publish a fact without a source behind it.
  • Silently discard extracted data because it was unfamiliar or hard to classify.
  • Present promotional or publisher-owned content as independent reporting.
  • Show a numeric "truth score" in place of real provenance.