How Was It True works

What gets checked

Factual assertions about companies, products, markets and technology, said by an identifiable speaker, where the words can be quoted and linked. A claim has to be falsifiable in principle and settled by documents that already exist. It also has to answer a specific reader question about a material decision, a meaningful discrepancy, or context that changes how the claim reads.

A routine restatement of a filed metric does not get its own check merely because it is factual. Every published check states why it was selected.

Not checked, at any evidence level: anything about a person’s character, conduct or motives; health and medical claims; elections and partisan politics; predictions; opinions and value judgements.

What counts as a source

Every consulted source is classified into one of four tiers, strongest to weakest: Primary, Official, Secondary, and Weak. The tier travels with the quote onto the page, so a reader can weigh the evidence the way the pipeline did.

Primary.Primary documents only: regulatory filings, official transcripts, government statistical series, court records, standards publications, peer-reviewed papers, and a company’s own publications about itself. Reporting on a source is not the source.

A claim needs two independent such sources. Independent means a different publisher, a different underlying document, and neither citing the other — so a company’s own filing and that same company’s press release count once, not twice.

There is one exception. A claim about a company’s own reported figures may rest on a single source when that source is a periodic report the company filed with a securities regulator — a 10-K, 10-Q, 20-F, 40-F, 8-K, 6-K or DEF 14A. Only the company knows its own quarterly revenue, so a second independent source for it does not exist. What makes one filing enough is not that the company said it twice; it is that named officers signed it under Sarbanes-Oxley, with liability attached. A press release is not eligible however similar its contents.

Official.A first-party statement that is not a primary source about its own subject: a company’s announcement about a deal with another company, a government or regulator press release, an official investor-relations page, a named executive’s own published statement. Signed by someone with standing to know, not attested under liability.

Secondary. Reporting by a named outlet with an editorial process and a corrections policy: a wire service, a newspaper or magazine of record, a trade publication with a masthead, a court or regulator reporter. The report must name its own sourcing; a report that cites another report is that report, not a new one.

Weak. Everything else: personal blogs, forums, social posts, aggregators, encyclopaedias, content farms, and any page whose author or editorial process cannot be identified. A Weak source never counts toward any floor and never appears beneath a verdict, on its own or alongside a stronger tier. It may be recorded in the research trail as a lead that was followed.

Independence across tiers follows the same rule as independence within a tier: not the same publisher, not the same underlying document, and not one citing the other. Two outlets reporting the same press release are one Official source and zero Secondary ones.

A claim that cannot clear the two-Primary-source floor may still publish on the open-research path when at least two independent sources of tier Official or Secondary bear on it, each with a quoted passage and a URLthat resolves to that passage. A Primary source always outranks every lower tier: it counts toward the two, and a conflict between a Primary source and a lower tier is resolved in the Primary source’s favour, never rendered as a contested verdict. The page states its evidence basis — the highest tier the verdict rests on — beside the verdict, and the research trail behind it — every query run, every page fetched, and every lead kept or dropped, and why — is recorded with the claim and reviewable.

Quoted words, and why the wording is stated

An automatic caption track is how a claim is found. Automatic and automatically translated captions are admissible attribution with explicit provenance. A claim taken from either track may publish without separate human confirmation. The claim records automatic_captions as its wording provenance; missing or unknown provenance still fails closed. A claim may publish only after its attributed timestamp is unchanged and the primary-source, independence, and verdict floors clear. Caption models fail on exactly what this site checks — numbers, units, scale words and product names — and a mis-heard “800 million” published as “800” would be a false claim attributed by name to a real person. Every page says its wording provenance. Caption errors are resolved by the ordinary evidence and verdict process rather than by a separate wording gate.

The verdicts

Supported: The primary sources agree with the claim as stated.
Unsupported: The primary sources do not bear the claim out. This is a statement about the evidence, not about the speaker's knowledge or intent.
Contested: Qualifying primary sources genuinely conflict with each other. Both sides are shown.
Missing context: The claim is accurate as far as it goes, and a fact in the sources materially changes how it reads. That fact is quoted.
Outdated: The claim was supported when it was made and the sources have since moved. Both the then and the now are shown, with dates.

Absence of evidence is not evidence. Unsupported requires a source that bears against the claim: a figure that differs, a record that says otherwise, a date that does not match. A search that finds nothing on the point, or finds only figures that answer a different question, yields no verdict and no page. Contested means the sources conflict, not that the evidence is thin.

A verdict is a summary of the sources printed beneath it and nothing more. It is never a statement about what a speaker knew, meant or intended. There is no accuracy score for any speaker on this site, at any aggregation: the subject of a verdict is a claim, never a person.

We publish selected claims whether they check out or not. The editorial bar is reader value, never a preference for negative verdicts.

Corrections

If a check is wrong, say so here. A report is treated as a defect in the pipeline, not as feedback: the claim comes down in the same change that fixes whatever let it through, and a page that was wrong and is now right says so, with the date.

Every correction made so far is listed on the corrections page.