Submit

A paper is an argument, and it will stay one. But the reusable part of most research is not only in the prose but it is the method, the code and the decision rules polished for the submission. Relying only on the prose leaves that part locked in a PDF for other people to reconstruct by hand.

What changes

When the initial consumer becomes and LLM acting for a researcher, the question shifts from “is this written clearly?” to “can this be used correctly?” Making your methodology callable the way you intend multiplies who can build on it: no more cloning repositories and reimplementing code, other researchers can hand your agent their data and get the answer.

One source of truth

Today, reusing a method most of the time means cloning a repository and editing it. Every clone is a chance to introduce a mistake that the original authors never made and will never see, and the resulting divergence is invisible: a hundred slightly different versions of one method, all citing the same paper.

A Paper Agent removes the fork. There is one implementation, the authors’ own, and it is the one that is executed. When it needs to transform someone’s data to do its job, the transformation is part of what it reports rather than something buried in a local script, so the person using it can see what was done to their numbers.

Where we stand today

Third-party submission is not yet built. The catalogue is curated by hand for development purposes, and every record in it was entered by us. What already exists is the part underneath: five works whose agents run their papers’ real code or packages.

See what using a work looks like today →