How our stories are made and checked

AgingResearch.News reads every new paper in PubMed, bioRxiv and medRxiv, new biology and biomedical computing papers on arXiv, new papers on Research Square and other preprint servers, the journals whose articles no longer reach PubMed, and the aging and geriatrics articles of Japan’s journals on J-STAGE.

Each paper is read for six questions: how relevant its own work is to aging, what link to aging that work shows, how newsworthy it is, what kind of study it is, what it mainly establishes, and whether it is promotional.

A paper gets a story when it passes four tests: it is primary research; the judging model finds it more likely than not relevant to aging, with the link shown in its own work; it scores at least 1 out of 3 for newsworthiness; and it is not promotional. A well-run randomized trial in people about aging gets a story whatever its newsworthiness score. Reviews get none.

An AI writer writes each story from the paper’s abstract and, where the paper has full text, excerpts of its methods and results, in plain words for a reader outside the field. When it cannot, another AI writer writes the story from the same text under the same rules.

The judging model then scores the story against that text, from 0 (contradicts the paper) to 3 (fully faithful). A story below 1.9 is not published.

A second model then reads each story that passes for claims that go further than the paper: about what kind of paper it is, how the study was done, which organism or population each result came from, and why it matters. A story it flags goes back to the writer, and a story it still flags for claims beyond the paper is not published.

When a draft calls a detail of the design missing (how long, how many, the dose, the ages, the sex, the follow-up) and AgingResearch.News holds the paper’s methods and results, they are searched for it, and a draft that calls a given detail missing goes back to the writer. With the abstract alone, a story may say only that the abstract does not give it.

Figures are the papers’ own, shown with their credit and license; other pictures are posters drawn for AgingResearch.News.

A story shown to misreport its paper is corrected or withdrawn. When a paper’s authors point out an error, the story is corrected.

Every correction is on the corrections page with its date. To report an error, write to us from the contact page.

Boris Djordjevic, the founder, is the editor of record: he answers for these standards and for every correction.

More about Boris Djordjevic and who publishes AgingResearch.News is on About.

Subject pages show where the evidence on a subject stands in one line. How to read it:

• C, W, M, H: cells, worms or flies, mice, humans. Dark: independent labs found the same change there. Light: one lab or one study. Outline: nothing yet.
• Labs are independent when their papers share no last author and no institution.
• Against: findings that point the other way from most findings on the same outcome.
• About aging: the drug’s trials about aging. The total counts its trials for every use.
• No results posted: trials that ended over a year ago with no results on ClinicalTrials.gov and no paper reporting them.
• Company ties: findings with an author at a company, or from a paper whose competing-interests statement declares ties to one.