Identify checkable claims
The system looks for concrete statements about people, events, numbers, policies, dates, and places.
Methodology
How FactScope reaches a result—and where it stops.
FactScope organizes evidence for investigation. It does not know ground truth, replace the original source, or guarantee that every checked statement is accurate.
The analysis pipeline
A page scan passes through several distinct stages. Keeping these stages separate prevents a reputable publisher, polished writing, or a single related article from becoming proof that every claim is true.
Understand the content type
Opinion, satire, predictions, breaking news, and pages without checkable claims receive different handling.
Find candidate sources
Fact-check results and news reporting are searched separately for each extracted claim.
Check evidence quality
FactScope evaluates reachability, relevance, recency, independence, duplicated reporting, and self-corroboration.
Separate evidence from context
Direct findings, matching coverage, and broader background appear in different groups.
Keep uncertainty visible
Limitations and technical failures are shown instead of being converted into a false low-trust result.
What the labels mean
| Label | Meaning | What it does not mean |
|---|---|---|
| Supported | Qualifying direct evidence or strong independent corroboration supports the checked claim. | Every detail in the page is proven. |
| Contradicted | Qualifying evidence conflicts with the checked claim. | The publisher or entire page is always unreliable. |
| Mixed | Evidence is contextual, nuanced, or points in different directions. | The claim is exactly half true. |
| Matching coverage | Reporting appears to discuss the same claim or event. | Every detail has been directly confirmed. |
| Broader context | A source provides useful background but is not a close enough match. | The source supports or contradicts the claim. |
| Still open | No qualifying external evidence was found within the scan. | The claim is false. |
How confidence works
Confidence describes how strongly the available evidence supports a specific FactScope finding. It is not the probability that a claim is true.
Direct and inspectable
Normally reserved for a reachable direct fact-check whose reviewed claim closely matches.
Strong corroboration
May come from qualifying independent reporting or an authoritative primary source.
Not enough for a finding
Used when evidence is unavailable, indirect, still developing, or cannot be inspected strongly enough.
A different measurement
Shows how many claims have matching reports. Broad coverage can still leave exact details unresolved.
Image verification uses separate assessments
- Visual manipulation indicators look for signs associated with generation, compositing, or editing.
- Caption consistency compares checkable caption claims with available external reporting.
- Visible source clues identify credits or watermarks that may help trace provenance.
Important
Visual analysis is model-based, not cryptographic or forensic verification. Editing can be legitimate, and a visible credit does not prove ownership or an unedited chain of custody.
Source and presentation signals
HTTPS, author attribution, publication date, publisher history, layout, and risky presentation patterns help users understand a page. These signals remain secondary and do not establish whether individual claims are true.
Known limitations
- Breaking-news evidence may be incomplete or change quickly.
- Search results can miss local, paywalled, newly published, non-English, or poorly indexed reporting.
- Matching headlines may describe a different time, place, person, or event.
- Syndicated reports are not necessarily independent confirmation.
- AI models can misclassify content or overlook relevant details.
- FactScope should not be the sole basis for medical, legal, financial, safety, or civic decisions.
For consequential claims, open the linked sources, find the primary material, compare dates and wording, and consult a qualified expert when appropriate.