The case for no gates.

A paper that enters the scientific record gets cited, built on and taught. If it's wrong, the error travels with it. So journals put expert reviewers between an author and the reader, and preprint servers put moderators at the door. The record had to be right.

There were never enough reviewers to handle it all. In 2020 alone, researchers gave more than 100 million hours to peer review, unpaid.¹ Papers wait months for a referee. And the review itself has become someone else's profit: Elsevier, the scientific publishing business of RELX, reported £1,035 million in adjusted operating profit for 2025, about $1.4 billion.² Its journals run on review that experts give for free.

Picture a reader holding a paper by someone they've never heard of. Until recently there was one cheap question they could answer: where was it published?

The journal's name stood in for everything the reader couldn't check. Had anyone qualified read it? Did the methods hold up? Did this author have a history of getting things right?

Nobody could answer those questions one person at a time. So the journal answered them for everyone, and a handful of journals and servers came to decide whose work counts.

Two things changed.

The evidence is open. The Retraction Watch database, more than 67,000 retractions, became free to everyone when Crossref acquired it in 2023, and it's updated every working day.³ The links between papers are public. Who built on whose work, what was corrected, what was withdrawn, whose results held up: the track record of a paper, and of the people behind it, is now a matter of public record.

There is an intelligence that can read all of it. No editor can hold a whole field's citations in their head. A machine can read every paper, check every reference against every retraction notice, and follow a claim through everything that cites it.

Pith puts the two together. We read the papers, map their citations, check each reference against retraction notices and publish all of it beside the paper. The signal a journal's name used to stand in for, Pith builds from the evidence itself, out in the open where anyone can check it.

Machines can read the record. The judgment that matters most is still a qualified person reading a paper closely. Pith pays for that judgment and publishes it.

Each review adds to the reviewer's public record: what they reviewed, the predictions they made and how those predictions turned out, and whether they tend to go easy. Over time that record is the credential. It belongs to the person, and anyone can read it.

More and more people meet science through an AI assistant. The assistant has to decide which sources to believe, and a journal's logo tells it very little. Evidence tells it a lot.

We built Pith for that reader from the start. Every review is published as structured data, and every current review ships in a monthly bulk download. A read-only server speaks the Model Context Protocol, the standard way assistants connect to outside tools, so an assistant can ask Pith for a paper's review, its claims, its signed human reviews, a reviewer's track record or the papers around it in the citation graph, and get the evidence back along with the judgment.

An AI that can check the evidence doesn't have to take anyone's word for it, ours included. That's the kind of source it will learn to trust.

A preprint is a paper shared before journal review, so the work can be read and argued with right away. arXiv made that normal in physics, mathematics and computer science.

For most of its 35 years arXiv was the closest thing science had to an open door. In the past year the door has narrowed.

The pressure behind these rules is real. arXiv receives about 20,000 submissions a month, screened by about 300 volunteer moderators, and much of what they reject is machine-written.⁷ arXiv calls the rise in non-scientific submissions “unsustainable.”⁵ But arXiv hasn't kept up with the tools that can now read papers at that volume, so it guards the door the old way. Its filter asks who you are and who will vouch for you before the paper can even be submitted.

A student with a correct result and no sponsor is now on the wrong side of that door. So is a researcher outside a university. A researcher who writes in another language now carries the cost of a full translation before anyone looks. A rule about credentials can't tell a strong paper from a weak one. Reading can.

Pith has one set of rules. They're published, and they apply to everyone the same way.

What Pith features is decided by the work and by the people who use it: citations from other papers, readers' votes and comments, and what the review finds the paper contributes. The author's name and institution never enter the score. Everyone gets the same seat, and the work decides how far it goes.

The journals' job was to keep the record trustworthy. That job matters more than ever, with machine-written papers arriving by the thousand.

What has changed is what we can do about it. We can read everything, show our work, pay the people who review, and let credibility build in the open, one paper and one person at a time.