Methodology & editorial policy

Crosscheck has no view on the news. It has a view on how news should be presented to you: attributed, plural, and checkable. This page explains exactly how the edition is made — including where it is weak.

1. What this site is (and is not)

It is an aggregator with a spine. Crosscheck collects the RSS feeds that 38 newsrooms already publish, groups headlines that describe the same event, and presents each group with a coverage bar showing how many left-, centre- and right-rated outlets ran it.

It is not “neutral” by pretending to have no perspective. Choosing which outlets to monitor, and how to rate them, is a judgement. We publish that judgement in full — the source list and every rating are below, and the code that does the grouping is in the repository. You can disagree with our inputs and still use the output.

Balance is a property of a set of headlines, not of a single sentence. That is why we show you the set.

We do not write articles, we do not summarise them, and we never paraphrase a headline. Every headline on the site is the exact string the outlet published, rendered in a serif face and attributed in the line below it. If an outlet is wrong, the reader can see that it is wrong, because the original is one click away.

2. How a story is built

1. Collect. Every 30 minutes the build fetches each outlet’s public RSS feed (the same feeds their own site uses). No scraping of paywalled pages, no circumvention of meters, no full-text copying. We keep headline, link, publication time and the short feed summary — the summary is used only to help group headlines and is not published.
2. Group. Headlines are converted into weighted token vectors (TF-IDF), with proper nouns boosted, and clustered around centroids. Two headlines join the same story only if they clear a similarity threshold and share distinctive terms — so “Senate race in Michigan” and “Senate race in Alaska” stay apart. Clusters need at least two different outlets to be published.
3. Label. Each outlet carries a lean rating (§3). The coverage bar, the blindspot flag and the word-choice line are all derived from those ratings.
4. Present. One representative headline is chosen — preferring centre-rated outlets, then the headline closest to the cluster’s centre of gravity — purely so the card reads cleanly. All other headlines are one click away, unedited.

Word choice watch lists the terms that appear in the headlines of only one side of the spectrum within that specific story. It is a prompt to look closer, not a claim that an article is biased. A single word is not a worldview.

3. How outlet lean is rated

Ratings describe a publication’s average editorial position over time, on a five-point scale: Left, Lean Left, Centre, Lean Right, Right.

English edition: ratings are third-party assessments — principally the AllSides Media Bias Ratings, cross-checked against Ad Fontes’ reliability/bias work and the Reuters Institute’s Digital News Report — not our own scoring of individual articles.

Arabic edition: no comparable public rating body covers Arabic-language outlets at scale, so Arabic ratings are Crosscheck’s own provisional assessments. They rest on three disclosed inputs: ownership and funding structure; a documented editorial line (who the outlet endorses, whose wars it justifies); and the outlet’s framing of three reference events. They are weaker evidence than the English ratings and are marked as provisional in the table below. Corrections are especially welcome here.

• A lean rating is not a truth rating. A Left-rated outlet can be first and right; a Centre-rated outlet can be wrong. Lean describes editorial posture, not accuracy, and we publish no reliability score alongside it so that it is never read as one.
• Outlets are rated, not articles. A single op-ed may argue the opposite of its paper’s line. Opinion pieces are marked “opinion” wherever we can detect them.
• Ratings are reviewed, and can be disputed. Send examples; ratings change with the record, and a change is noted here.

We deliberately include low-reach, ideologically explicit outlets alongside legacy wires. Removing them would make the bar look balanced while hiding the extremes that shape the argument.

4. Blindspot detection

A story is flagged as a blindspot when, in an edition where at least three outlets covered it, one side of the spectrum is entirely absent and the other side carried it with at least two outlets. This is deliberately conservative: a thin two-outlet pair is “thin coverage”, not a blindspot, and the flag is not applied to it.

Absence of coverage in our feed list does not prove absence of coverage everywhere — some outlets publish nothing in RSS, and a newsroom may have covered the story on a platform we do not read.

5. Full source list

38 outlets in this edition’s language, 31 of which published a story we could cross-check. The number in the fourth column is how many headlines that outlet contributed.

To propose a source, email the editor with its public RSS URL. The bar for inclusion: a working public feed, original reporting, and enough of an identity to rate. State-funded outlets are included only with a label; state propaganda operations under sanctions are excluded.

6. Funding & independence

Crosscheck is built to run for almost nothing: static files, no database, no reader accounts, no advertising, no venture money. It currently has no revenue and no sponsors, and it is not funded by any political party, campaign, government or advocacy group.

• No advertising and no sponsored content. Outbound links are plain links — no affiliate parameters, no referral tracking.
• No personal data. No cookies, no analytics scripts, no fingerprinting, no comment system, no newsletter signup that harvests addresses. The host sees IP addresses in its access logs, as every web host does; we do not build profiles from them.
• If money arrives later, it will be disclosed here before it is spent: reader donations or grants, never paid placement, never a stake in coverage decisions.

7. Known limitations

• Headline-level. Grouping reads headlines, not articles. Two outlets can share a headline and mean different things; two can differ in wording and agree completely. The comparison view exists so you can judge that yourself.
• RSS is an incomplete sample. Many large outlets (Reuters, AP, The Washington Post at times, most Gulf outlets with Cloudflare-blocked feeds) publish no usable public feed, so they are absent — and their absence skews the balance bars.
• Five-point labels flatten real variation. Within “Centre” sit a wire service and a magazine with a strong worldview; within “Right” sit libertarians and nationalists who disagree with each other more than with the centre.
• Clustering is imperfect in both directions. It occasionally merges two related stories or splits one. Grouping is deterministic and reproducible; it is not infallible.
• Two languages only. English and Arabic. Hebrew, Spanish, Hindi and Turkish coverage of the same events is invisible here.
• No fact-checking. We do not adjudicate who is right. Labelling a claim’s origin is our job; verification is not a job we have staff for, and pretending otherwise would be the most dangerous thing this site could do.

8. Corrections policy

• Misgrouped story? Email the story URL and the headlines that do not belong. Misgroupings are logged and the clustering thresholds are adjusted.
• Misrated outlet? Send at least three examples across at least six months. Ratings are reviewed on the record.
• Broken link or missing feed? Reported within a day; the outlet is fixed or removed.
• Serious factual error in our own copy: corrected within hours, with a dated note kept at the bottom of the page — never silently edited.

9. Contact

Editor: editors@crosscheck.news. We answer corrections before anything else. There is no comment section by design: the argument about the news belongs in the outlets we link to, not in a moderation queue we cannot staff.

The table

Edition built 2026-11-01T21:45:11.716Z · 47 · 138 · 31