Methodology.
How a claim becomes a score, and how a score gets published.
METHODOLOGY v4.2 · LAST REVISED JULY 2026
The score is a summary, not a verdict
We don't publish TRUE or FALSE. Evidence rarely sorts that cleanly, and a binary label hides more than it reveals. Instead every check gets a credibility score from 0–100 with a confidence interval — for example 72 ±6 — and that interval is honest: it's the range the evidence actually supports, not a hedge to cover our own uncertainty.
The score is computed from four visible dimensions, and the breakdown behind each one is always shown alongside the number it produced:
Sourcing
Are the claims backed by primary or independently verifiable sources, and how many hops removed is the coverage from where the information originated?
Factual accuracy
Do the specific figures, dates, and claims match what those sources actually say — not just whether the general gist is in the neighborhood of true.
Framing
Accurate facts can still mislead through selective emphasis, missing context, or a headline that implies more than the body supports. We weigh that separately from raw accuracy.
Provenance
Who published this, why, and what track record and incentives sit behind it — including whether content is sponsored, syndicated, or promotional in nature.
What we don't score
Some content isn't a factual claim to begin with, and scoring it as one would be its own kind of dishonesty:
- Satire and clearly-labeled parody are identified, not scored — a joke isn't a claim, even if it spreads like one.
- Opinion and editorializing are noted as opinion. We check the facts an opinion piece relies on, not whether we agree with its conclusion.
- Promotional content isn't automatically "fake" — sponsored posts and ads are flagged as promotional, and evaluated on their own terms, not penalized simply for being commercial.
Every published check passes editorial review
Analysis is AI-assisted, but nothing reaches this archive automatically. A contributor submits a check, an editor reviews the sourcing and reasoning behind it, and only then is it published under that editor's accountability. If a reviewer sends a check back, the contributor sees exactly what needs fixing before resubmitting.
Scores can move — and we show it
New evidence sometimes surfaces after a check is published. When a score is revised, the report keeps a visible history of every change and the reason for it, rather than quietly overwriting the old number.
What we keep to ourselves
We publish our principles and our process in full, on this page. We don't publish the exact prompts, models, or weighting formulas behind the scoring engine — the same way a newsroom explains its editorial standards without handing out its internal style guide verbatim. That's a deliberate line: enough transparency that you can evaluate whether to trust a score, not so much that the methodology becomes a manual for gaming it.