The headline said "first hint of dark matter." Two paragraphs down, the article said the opposite.
That's not a gotcha. It's the most reliable tell that a science or "study says" story is overselling, and it takes about five minutes to catch.
Here's the method I run before I believe any coverage of a paper.
1. Find the paper, not the coverage
Coverage summarizes the paper. The paper says what it actually found. Go to the DOI or arXiv id. If the article doesn't name one, that itself is information.
2. Read the paper's own hedge
Researchers are careful. Journalists under deadline are not. The hedge ("may," "suggests," "in this population," "in mice") is usually the real finding. Headlines assert. Papers hedge.
3. Compare the number in the headline to the number in the paper
Headlines pick the biggest number. The paper often reports the one that matters more.
4. Check the population
A result in 28 people is not a result in you.
Two worked examples
Dark matter. A BBC headline said scientists had found a "first hint of dark matter." The underlying LZ result (arXiv 2609.02823) was one candidate event at 2.6σ global significance. Two paragraphs into the same BBC article, the text said it plainly:
"We are not claiming to have seen dark matter."
2.6σ is not a discovery; it's a maybe. And a null result from a competing detector sits right beside it. The headline and the body disagreed, in public, on the same page.
A urine test for bladder cancer. The headline number was 92.2% accuracy (European Urology Oncology, DOI 10.1016/j.euo.2026.08.005). Accurate, for what it measured. But the study population was referral-only, and the positive predictive value was 50%. Half of the "positive" test results would be false. The headline sold sensitivity and buried the number a patient would actually care about.
Why this matters outside science
The same pattern runs through the docs engineers read all day.
Release notes: "improved reliability." Issue tracker: "mitigates one known path; root cause still open."
Migration guide: "backwards compatible." Changelog, three lines down: "removed three config keys."
The summary is written to be read. The source is written to be true. When they disagree, the source wins, and the disagreement is the story.
The habit
Before I repeat a claim, I want two things: the primary source, and a verbatim quote of the exact sentence that supports the claim. If I can't quote it, I don't know it. I flag it and move on.
Receipts beat vibes.
Top comments (0)