From your feed to the public record
Four signals per outlet, aggregated from trust-weighted reader votes, with distribution visibility and brigading defenses. No editor in the loop.
3 min read · The full detail is in the methodology1. Readers leave reviews
Star rating + a bias pick on five steps + an accuracy call (Accurate, Mixed or Misleading) + one sentence. Quick path takes under 30 seconds. Detailed reviews (dimension ratings, evidence URLs, pros/cons) are opt-in.
2. Votes are trust-weighted
Your vote weight scales with review history quality — outlets reviewed, helpful votes received, consistency across the bias spectrum. New accounts count less than long-tenured reviewers.
3. Distribution is public
Every outlet page shows the full vote histogram, not just the median. Polarized outlets look polarized. Consensus outlets look consensus. Outliers self-disclose.
4. Raids are capped and frozen
A new verdict waits 36 hours before it moves a published score. Anonymous votes from any one day can add at most a quarter of the weight a record already has, and when the brigading check sees coordinated voting it freezes the score at its pre-incident value — the record says SCORE FROZEN — until a moderator reviews it. Sustained shifts require sustained crowds, not pile-on raids.
The four signals per outlet
Overall trust score (1–5 stars). The average of jurors’ star ratings, weighted by each juror’s track record, with recent verdicts counting more. Published once an account has 20 ratings.
Political bias (seven published steps).Far Left, Left, Lean Left, Center, Lean Right, Right, Far Right — the weighted median (not mean) of jurors’ five-step votes, for brigading resistance. Confidence levels: Hidden (<20 votes), Low (20–99), Moderate (100–499), High (500+).
Accuracy score (0–100). Reader-aggregate “how often does this outlet’s factual claims hold up under verification?”. Normalized to 0–100. Labels: Unreliable, Questionable, Mixed, Mostly Accurate, Highly Accurate.
Crowd tags. Community-submitted labels (clickbait, well-researched, well-sourced, opinion-heavy, etc). Each tag appears publicly only after crossing a voting threshold.
Why crowd-sourced beats editor-led at scale
A 5-editor team (AllSides) can rate ~600 outlets in 10 years. A single-researcher operation (MBFC) can hit ~5,000 with consistency questions. A paid journalist staff (NewsGuard) caps at ~10,000 behind a paywall.
Crowd-sourced ratings with proper defenses scale without those caps. The trade-off: brigading is a real risk. Our defenses (trust weighting, distribution visibility, a daily cap on anonymous votes and a freeze when a brigade is detected) are documented in the methodology page and explored in depth in this post.
What we don’t do
- We don’t fact-check individual articles. We tell you whether a source has been reliable; not whether this story is right.
- We don’t allow AI-generated reviews. AI is used only for spam + toxicity detection on submissions.
- We don’t hide unfavorable distributions. If an outlet’s reviewers are polarized, it shows on the page.
- We don’t remove negative reviews for substance. Only policy violations (impersonation, off-topic, demonstrably false claims about the outlet).
- We don’t sell ratings. An outlet that claims its page can publish a response — it does not affect the score.
Ready to use it?
Three ways to start: browse the bias map, take the 2-minute quiz, or paste a news URL on the rate page.