Our method

How Paradox News finds, rates and presents the news, and its limits.

The lens: hold the opposites

Most news picks a side or splits the difference. We do neither. For each big story we try to understand both sides before anyone judges:

• Steelman each side. We give its strongest honest case, in its own best voice, with real quotes and sources.
• Name what each side is protecting. This is the value or fear underneath, put in words the other side would accept as fair.
• Set out the common ground and where the evidence sits. Each disputed claim is checked against primary sources and marked supported, contested or unsupported.
• Show what each side leaves out. Every side has inconvenient facts.
• State the paradox. Both things can be true at once. We often leave the tension unresolved.
• Offer the bridge. This might be a shared question, a both/and framing, or the evidence that would change each side's mind.

We never cheerlead any party or campaign. Every party, company, union, charity and campaign group gets the same scrutiny.

Topics and tagging

Stories are grouped into 14 topics: Global, Israel, UK, UK politics, Greens, Political economy, Business & economy, Technology & AI, Science, Health, Climate & environment, Ecology & land, Culture and Sport. For the new subject topics we read section feeds from left-, centre- and right-rated outlets so no single lean decides what counts as news. Each story's first tag comes from the section it was chosen for. Further tags come from a fixed keyword list matched against its headline and search terms, so one story can sit under several topics (a bank cutting jobs because of AI is both Business and Technology). The rule is the same every day and we don't hand-pick tags. Every topic story gets the same treatment as the rest: equal articles per side, blindspot flags and Follow the money. Browse them on the [Topics](/topics/) page.

How we measure bias

Every outlet is rated using Media Bias/Fact Check (MBFC). Each page is checked to make sure it matches the right outlet. Five ratings become three groups: Left and Centre-left count as L, Least biased counts as C, and Centre-right and Right count as R. Each story's bar shows the share of rated outlets on each side that covered it in our sample. Outlets without a rating we can verify are listed as "Not rated" and left out of the bar. See every rating on the [Sources page](/sources/).

Blindspots

We test the left and the right in exactly the same way, each against its own usual share of our sample (about 50% L, 15% C and 35% R).

• The test. Say a story is covered by n rated outlets. We work out the chance that the left (or the right) would get as few of them as it did if outlets picked stories at random from our table. We flag a blindspot for that side when that chance is 10% or less.
• At least four rated outlets. We don't flag stories with fewer.
• Both sides are checked separately. Neither side is checked first, so a story can be a blindspot for both.
• Display. The Blindspots page and the email show the two sides in parallel columns with the same heading and a count each. The side with more blindspots goes first; on a tie the order alternates by day. If a side has none, we say so and why, rather than filling the column.
• Centre-only / one-outlet stories are listed separately.

Why we changed it (23 Oct 2026). The old rule flagged a side when its share fell below half its usual share. Because the right has a smaller share of our sample, that bar was far easier to clear for the right than for the left. For a 12-outlet story, a right blindspot fired on a result with a 14% chance of happening randomly, but a left blindspot needed one with only a 2% chance. The old rule also checked the right first. The new test uses the same chance level for both sides.

A blindspot isn't automatically a scandal. It shows what each audience isn't hearing.

Balanced article lists

For every story we show the same number of articles from the left, the centre and the right, up to three from each side. We pick each side's most on-topic articles, preferring pages that open without a paywall.

• A short side. If one side has fewer articles, we first search that side's outlets again for real coverage of the same story. Anything we add this way is marked top-up and is not counted in the bar. If a side is still short, every side is cut to match it, so you might see 2 from each side, or 1.
• No coverage on one side. We show the other two sides equally and say plainly: "No centre coverage found — blindspot" (or left, or right). We never pad a side with unrelated articles.
• The bar is not balanced. The bar and the L/C/R counts always show the true totals from our sample. Under each list we say how many we're showing out of how many outlets covered the story, with a "More coverage" link to everything else.

Limits. Ratings are one organisation's judgement, and they change over time. We recheck them monthly. Our outlet sample isn't a census, so counts depend on what our searches find in a 48-hour window. Paywalled pages are listed but not quoted. Left and right are a rough axis and leave out a lot.

Where the news comes from

• News outlets. We draw on about 90 UK, US, Israeli and international outlets. We use their RSS feeds, plus Google News searches aimed at each part of the spectrum, so that right- and left-leaning outlets both get a fair chance to appear. Every link goes to the original article, and we never republish it.
• X. We record post counts per story, and X's own story summaries. X's summaries are AI-generated and can be wrong, so we label them and never treat them as fact.
• Reddit. We list public thread titles where they match a story. Subreddit bias is not rated.
• Summaries are our own, neutral and based only on facts the sources agree on.

Follow the money

Every story has a "Follow the money" panel. It asks who might gain or lose financially, and who funds the people and outlets driving the story. We frame these as possible interests and questions worth asking, never motives. A financial interest doesn't make an argument wrong.

• What we look at: who gains or loses from each outcome; who owns or funds the outlets covering the story; and the funding and lobbying behind the main voices on every side. That includes companies, unions, NGOs and charities, think tanks, donors and parties, including the Greens and environmental groups.
• Sources: Companies House, the Electoral Commission donations register, lobbying registers, annual reports and charity accounts, InfluenceMap/LobbyMap, OpenSecrets (US) and Transparify (think-tank transparency). We name the source when it is a campaigning outlet.
• Rules: real sources only. Where we've looked and found nothing, we write "no funding data found"; where we haven't looked yet, we write "not yet researched". We never guess. We name no conspiracy and assign no motive.
• Most stories show outlet ownership (from MBFC) and are marked not yet researched until we do a deep dive.

Corrections

If we get something wrong, we fix it and say so on the story. Please tell us using the address on the subscribe page once digests launch.