// METHODOLOGY

How a Fragment Zero case file is produced.

Topic selection

Cases are selected on three criteria: (1) the event is documented in public records — court filings, indictments, declassified intelligence reports, peer-reviewed publications, or named primary sourcing; (2) the event has not been adequately explained by mainstream coverage in a single accessible long-form treatment; (3) the event is consequential — it altered policy, infrastructure, or institutional trust in measurable ways.

Source hierarchy

Fragment Zero ranks evidence in the following order, from strongest to weakest:

1. Court records, sworn affidavits, indictments, prosecution memoranda.
2. Government documents released by request or declassification — FOIA, CRS reports, OIG reports, congressional testimony transcripts.
3. Peer-reviewed academic publications.
4. Investigative reporting by established outlets — Washington Post, New York Times, Wall Street Journal, Reuters, ProPublica, the Miami Herald — particularly where independently confirmed across multiple outlets.
5. On-the-record statements by named primary sources.
6. Technical disclosures by vendors (CVE advisories, FBI Flash alerts, CISA advisories) where the disclosing party has direct knowledge.

Anonymous-source claims, unverified leaks, and speculative reconstructions are not used as the foundation of any specific factual assertion in a Fragment Zero case file. They may be acknowledged as context, but the factual spine of the narrative is built only on records that can be cited.

Research process

Each case file undergoes:

• A primary-source review (the underlying court records, indictments, technical advisories, peer-reviewed papers).
• A literature pass over major investigative reporting on the topic, with attention to retractions, corrections, and disputed claims.
• A timeline reconciliation: events are placed on an absolute timeline and cross-checked against any source's reported chronology.
• A claim-by-claim verification pass: each factual statement in the narrative is mapped to its source. Statements that cannot be sourced are removed.
• An adversarial review: known critiques of the consensus reading are surfaced and addressed in the narrative when relevant.

Production pipeline

Once research is complete and a script is written by the editorial team:

• Voiceover is synthesised by a text-to-speech model. The voice is derived from a recording the editorial team is licensed to use.
• Visual illustrations are generated by image-synthesis models from prompts tied to the underlying narration window. They are stylistic interpretations, not photographs of the events described.
• Archival clips are sourced from public repositories — the Internet Archive, Wikimedia Commons, the Library of Congress — under their respective licences and used in fair-use citation where appropriate.
• Subtitles in non-English languages are produced by a neural machine-translation model. Technical terms (CVE identifiers, organisation names, place names) are spot-checked for accuracy.
• The final video is rendered, reviewed by the editorial team for sync and factual integrity, and published.

Editorial control

No part of the AI-assisted production stack — voice synthesis, image generation, subtitle translation — has authority to introduce, modify, or remove factual claims. All claims in the published narrative are written by human editors and verified against the sources listed above. Where the editorial team is uncertain about a claim, it is omitted rather than asserted.

Updates and corrections

If new public information materially affects a previously published case file, an update note is appended to both the YouTube description and the archive page. Substantive errors are corrected within forty-eight hours of identification; the original assertion is preserved in the correction notice.

See also: /disclaimer · /about · /contact