Methodology
How we classify, how we weigh evidence, how we source, and how we use AI. This page is the contract; the database is held to it, mechanically.
The five categories
Every finding is classified into exactly one primary category. This separation is the product.
The recorded numbers do not reflect the votes cast.
Included when: Included only when a court ruling, an official annulment, or an authoritative observer or forensic finding establishes it.
Unequal media access, repression, illegal financing, vote buying, misuse of state resources.
Included when: Documented by election-observation missions, courts, or official investigations.
Cyberattacks, hack-and-leak, covert foreign financing, impersonation of media or institutions by a foreign state or proxy.
Included when: Government attribution, an indictment, a platform attribution, or two or more independent research attributions.
Troll farms, fake local-media networks, covert political ads, coordinated inauthentic amplification — foreign or domestic.
Included when: Platform takedown reports or credible research attribution.
Patterns in the data worth examining — explicitly not proof of fraud.
Included when: Peer-reviewed or methodologically transparent analysis, always published with its limitations.
The five confidence tiers
Every finding, attribution, and narrative carries exactly one tier. When in doubt between two, we use the lower one.
Standard: A final court ruling, an official government or parliamentary finding, or two or more independent authoritative investigations.
Standard: Credible multi-source research or reporting pointing the same way, without an official finding.
Standard: Presented as an allegation attributed to its source, never as fact.
Standard: Both sides are presented with their sources.
Standard: Refuted by a court, recount, audit, or forensic review. Kept visible as a narrative artifact.
The canonical example: United States, 2016
A Russian influence campaign, social-media operations, and hack-and-leak were documented by the U.S. Senate Intelligence Committee and the Intelligence Community. That vote-counting systems altered the recorded result was not established. A record you can trust holds both statements at full strength.
Sourcing rules
- Primary sources before journalism; journalism before commentary; commentary never as sole basis.
- Every external source is archived at the time it is cited, with the retrieval date recorded.
- No anonymous social-media post is ever the sole basis for a finding.
- We quote minimally and link maximally, summarizing in our own words.
- Every statistical claim links its method, its data, and its limitations.
AI transparency
Drafting and translation are AI-assisted, and we say so openly. AI is a drafting and structuring tool, never a citable source: every published fact must trace to a listed primary source. Machine-translated pages are labeled as such and point to the English source of record until a human reviews them.
Straight answers
- Does documented foreign interference mean an election result was changed?
- No. Interference in the campaign environment is a different question from whether the recorded vote count was altered. We keep the two strictly separate, and we say when the second is not established.
- Was the 2020 U.S. election stolen?
- No. Claims of a stolen 2020 election were rejected across dozens of courts, recounts, and audits, and are recorded here as debunked — held to the same evidence bar as every other claim.
- Do statistical anomalies prove fraud?
- No. Anomalies can justify examination, but on their own they are not evidence of fraud. We publish them inside a clearly marked limitations frame.