A fact check reaches a verdict through a procedure, not an impression. The steps that precede the rating do most of the work, and they explain why checks of the same statement can differ.
The checkable claim must be isolated first
Public statements mix verifiable assertions with predictions, opinions and value judgments. Only the verifiable portion can be checked.
Desks therefore extract a specific factual proposition, often rewriting a sentence into the narrow claim it contains, before any research begins.
Different desks may isolate different propositions from the same sentence, and that choice, more than the evidence, drives divergent conclusions.
Primary sources outrank secondary reporting
Checkers work toward the original document, dataset or recording rather than accounts of it, because summaries introduce compression errors.
Where a statistic is cited, the check locates the underlying series, the definition of what it measures and the period it covers.
Definitions matter disproportionately, since many disputes turn out to be two parties using the same word for different measured quantities.
The speaker is asked for their evidence
Standard practice is to contact the person or organization and request the basis for the statement before publishing.
This serves two purposes: it can surface a legitimate source the checker missed, and it establishes whether any basis exists at all.
A refusal to provide support is itself reportable information, and it is normally noted in the published check rather than left out.
Ratings encode more than true or false
Most desks use scales that separate accuracy from completeness, distinguishing a false claim from one that is technically accurate but omits decisive context.
Scales also account for claims that cannot be resolved with available evidence, which is a different finding from a claim being wrong.
Because scales are house-specific, a label from one organization does not translate cleanly to another's, and the written explanation carries the actual finding.
Corrections policy is part of the method
Checks are themselves subject to error, so credible desks publish their methodology and a corrections policy stating how mistakes are marked and disclosed.
Transparency about sourcing lets readers retrace the reasoning, which matters more than the verdict when a claim is genuinely contested.
It also allows disagreement to be specific, since a reader who accepts the sources but rejects the framing can identify exactly where the divergence begins.