How Findings works

Findings uses a different admission rule from the World Feed, and the reason is worth stating plainly, because it is the same reason most science coverage misleads.

Why counting outlets fails here

The World Feed publishes a story when several independent newsrooms report it. That test works for events. It fails for research.

A discovery reaches the public as one paper and one press release, usually written by the university’s press office under embargo. Ten outlets then publish within an hour of each other. Those ten are not ten newsrooms confirming a finding — they are ten rewrites of a single document, and most of them have not read the paper.

Applied to research, the corroboration filter would therefore be exactly backwards: it would score a thin press release with a good hook as strongly corroborated, while a serious paper covered by one specialist publication would fail to appear at all.

The rule this page uses instead

Findings counts nothing. An item is admitted only if it is a primary source and its identifier resolves:

Kind	Admitted on	What it means
Peer-reviewed paper	a DOI	published in a peer-reviewed journal
Preprint	a DOI or arXiv id	not peer reviewed — a claim, not a finding
Retraction	the DOI of the withdrawn work	the record correcting itself
Institutional record	the publisher’s canonical URL	a research body reporting its own instruments, missions, or data

Science journalism is not carried here. This is not a judgement about its quality — some of it is excellent, and much of it is more useful to a general reader than the paper is. It is a statement about what this particular page is for: the document itself, and a link that goes to it.

What we measured before deciding

The obvious shortcut is to take a science headline and look up the paper behind it. We tested that against Crossref, the scholarly metadata registry, on 2026-08-19.

Given an actual paper title, Crossref returns the correct DOI every time. Given a journalistic headline, it returns something confident and wrong:

Headline	What Crossref returned
“Pollen has a surprising problem — and honeybees have found a way around it”	a paper about superbugs
“Coffee drinkers have less fat, more muscle…”	a book chapter on meat
“Dark Stars may have left gravitational-wave echoes…”	an unrelated neutron-star paper

Crossref always returns something. Nothing in the result tells you it is wrong, and a DOI printed next to a headline looks like a citation whether or not it is one. Automated headline-to-paper matching would therefore manufacture citations that are false and look authoritative — a worse outcome than publishing no citation at all.

So identifiers are read from the publisher’s own feed and never inferred. Crossref is used for one narrow job: confirming that an identifier we were given resolves, and recording what it resolves to. Items whose identifier does not yet resolve are shown as not yet resolvable rather than quietly dropped — usually it simply means the DOI was registered in the last day or two.

Preprints

Preprints are carried and are labelled, everywhere they appear, as not peer reviewed. Excluding them would hide where a great deal of research actually appears first; presenting them alongside reviewed work without a label would be worse. A preprint is a claim its authors are willing to put their names to. It has not been checked by anyone else.

What this page is not

Not a measure of importance. It is ordered by recency within each kind, not by significance. A routine methods paper and a result that changes a field look the same here.
Not peer review, and not a check on the research. Confirming that a DOI resolves says the paper exists and is registered. It says nothing about whether the work is any good.
Not comprehensive. It reads a fixed list of journals, preprint servers, and institutions. Whole fields and most of the world’s non-English research literature are outside it.
Not a substitute for reading the paper. Every title here links to the source document, which is the point.