How the engine decides what deserves scrutiny

Political scope

The engine collects politics and public affairs: elections, government, laws, courts, immigration, economics, war, public policy, civil rights, education policy, public-health policy and politically consequential culture disputes. Shopping, ordinary sports, celebrity, lifestyle and entertainment are excluded unless government or policy is materially involved.

Balanced publisher discovery

Every monitored publisher is checked independently through official RSS feeds, alternate feeds advertised by the publisher, political section pages and publisher sitemaps. Extraction is balanced across publishers so a prolific outlet or a larger backlog cannot monopolize processing.

Claim worthiness

The extractor keeps no more than three consequential factual claims per article. Opinions, rhetoric, predictions, captions, bylines, biographies and trivial metadata are rejected.

Evidence hierarchy

Original laws, court records, official data, audits, full transcripts, filings and original research come first. Other reporting can locate or corroborate evidence but does not become truth through repetition.

Proof gate

Publication requires multiple reachable sources from independent domains, specific proof points, and either authoritative primary evidence or three independent reporting sources when primary records are unavailable.

Article misinformation star score

Every published article receives a visible 0.0–5.0 star misinformation-risk score. Five stars means the most serious evidence-backed problem; zero means the checked claims showed no material factual problem. This is not a popularity rating.

The article score is 65% the most severe verified claim and 35% the average severity of all checked claims in that article. Publishing more claims or more articles does not add points.

Worst Offender score

The score runs from 0 to 100 and measures rates, not volume. Each fact-checkable verdict receives a severity weight: accurate claims add zero, while missing context, exaggeration, cherry-picking, misleading claims, unsupported claims, mostly false claims and false claims receive progressively larger weights.

The final rate is 75% claim-level severity and 25% the average worst verified problem within each evaluated story. Bayesian shrinkage pulls small samples toward the sitewide average, preventing one or two articles from creating a sensational ranking. Provisional ranking requires at least 8 evaluated claims across 4 stories; full ranking requires 20 claims across 8 stories.

Article count never adds points. Opinions, rhetorical hyperbole and unverifiable or inconclusive results are excluded from scoring.

Automatic publication

The system operates automatically. Failed proof checks remain hidden and are retried later. Every public claim displays the evidence used, confidence, corrected wording and related claims from other publishers.

No automatic allegation of lying

A false claim is not automatically called a lie. Intent requires separate evidence that the speaker knew the statement was false.