Peer Reviewed reads everything published each week in a fixed set of journals, picks the ten stories worth your time, and writes them up from the papers themselves. One edition a week, plus a podcast and a PDF. Nothing is rewritten from a press release.
Every weekday we fetch what these 12 journals published, from OpenAlex: Science Advances, Proceedings of the National Academy of Sciences, Nature Communications, Science, Physical Review Letters, New Journal of Physics, eLife, Nature, Communications Earth & Environment, PLOS Biology, PNAS, Royal Society Open Science. That set is fixed and public — nothing gets in because it was trending, and nothing is added because it was interesting after the fact.
A typical week is around 292 papers. Retractions, corrections, editorials and duplicate records are dropped first, along with anything whose abstract is too short to judge. What survives gets a plain-language headline and a one-paragraph summary written from the abstract.
Those summaries — not the abstracts — are what the selector reads. Papers compete within their own field first, so a molecular biology week can't crowd out physics simply by publishing more. Field winners then compete across five sections with different tastes: an opening story, one that's relevant to daily life, one where the evidence is genuinely in debate, one that's strong on evidence, and one to close on. At most one section per field, so a single discipline can never take the whole issue.
Ten are written up in full, five more get a line each. Every paper we saw is listed in the edition, including the ones we discarded — you can see exactly where we stopped.
They're written by a language model, from the paper's own abstract and metadata, and then checked by a second pass that looks for claims the abstract doesn't support. Every story links to the paper by DOI. We never write from a press release or a news article, because those are where most overstatement enters.
Each story carries what most coverage leaves out: what was actually studied — mice, cells, a simulation, or people — the sample size when there is one, and the limitation the authors themselves put on the finding. If a result is in mice, the story says so in the first breath.
Some things we refuse to publish. When the writer declines a paper — usually dual-use work such as pathogen transmissibility — the run stops and a person decides. We don't route around a refusal by trying a different model.
We summarise; we don't referee. A summary is not peer review, and one study is not a settled question. The DOI is on every story for exactly that reason.
The pipeline marks a summary corrected when its checking pass caught
and fixed an overstatement before publication. Those are visible in the edition
rather than quietly patched — in a recent week, 37 of 542 summaries carried the
mark.
For anything that reaches you wrong: email hello@peerreviewed.app with the story and what's wrong with it. Factual errors are fixed in place with a dated note on the story, and the correction goes out with the next edition. If a paper we covered is later retracted, the story stays up and is marked retracted — deleting it would hide that we ever ran it.
One person, in the Netherlands. If it's useful to you, you can support it here — that's the whole business model for now. No advertising, no trackers beyond the cookie that keeps you signed in.