Every claim published on They Knew passes through a structured verification pipeline designed to eliminate speculation and ensure that only evidence-backed assertions reach our database. Below is a detailed breakdown of how the process works.
Every claim submission requires a descriptive title (minimum 20 characters), a context statement (minimum 50 characters), a category, and at least one primary source URL. Submissions lacking any of these are rejected at the form level. Submissions that are near-duplicates of existing claims are blocked automatically.
Every submission is scored 0–10 by Anthropic's Claude Haiku model against two criteria: (a) verifiability — can this be proven with primary sources — and (b) likelihood it is a real documented claim, not a random question or hoax. Submissions scoring below 5 are rejected with an explanation. Submissions scoring 5 and above enter the queue as Pending Review.
All new claims — whether submitted by users, curated by editors, or ingested from our scraper pipeline — enter the database as Pending Review. They are publicly visible with a yellow status badge that clearly marks them as not yet verified. This prevents any AI-generated or user-submitted content from ever appearing as "verified" without a human or community signal.
Registered users upvote, downvote, flag, and comment on Pending Review claims. A claim is automatically promoted to Verified (green) when it receives at least 15 upvotes from distinct users. Claims with 5 or more downvotes exceeding upvotes receive a visible Disputed badge. Claims can be flagged as false by any authenticated user; flag counts are displayed publicly for transparency.
A curated whitelist of historical classics — claims with decades of declassified documents, court rulings, or congressional hearings (MKUltra, COINTELPRO, Pentagon Papers, Operation Mockingbird, Tuskegee, etc.) — are marked Verified by editorial override. Moderators can also manually Approve, Dispute, or Debunk any claim from the admin panel with a recorded rationale. Every status change is logged and auditable.
Published claims are never considered permanently closed. New evidence can trigger a status review at any time. The flag system, community comments, and moderator queues provide continuous oversight. Our team periodically re-evaluates claims in active categories.
Sources are ranked by reliability. Higher-ranked sources carry more weight during verification. The hierarchy, from strongest to weakest:
| Rank | Source Type | Description |
|---|---|---|
| 1 | Government Admission | Official statement or report by a government body acknowledging the claim. |
| 2 | Declassified Document | Previously classified government or military documents released to the public. |
| 3 | Court Judgment | Ruling or finding from a court of law directly addressing the claim. |
| 4 | Congressional Hearing | Testimony or evidence presented before a legislative body under oath. |
| 5 | Peer-Reviewed Study | Research published in a peer-reviewed academic journal with reproducible methodology. |
| 6 | Investigative Journalism | Long-form investigation published by a credible news organization with named sources. |
| 7 | FOIA Document | Records obtained through Freedom of Information Act requests. |
| 8 | Expert Testimony | Statement from a credentialed expert in the relevant field, given publicly or under oath. |
Every claim on the platform is assigned one of the following statuses:
Every claim on They Knew includes a Conspiracy Viability Score based on the mathematical model published by David Robert Grimes (University of Oxford) in PLOS ONE (2016). The model calculates the probability that a conspiracy would be exposed based on the number of conspirators and the duration of secrecy. The core formula: L(t) = 1 − e−tNp, where p ≈ 4×10−6 per person per year. Read the full explanation →
They Knew uses large language models (LLMs) as part of its research and content pipeline. Here is exactly how and where AI is involved:
