Sources & scope
The paper, equations, assumptions and limits of the check.
RESEARCH
From published claim
to reproducible result.
We check specific derivations, numerical results and observational constraints in astrophysics and cosmology papers, mostly from arXiv. One narrow claim at a time, not whole papers.
PUBLISHED ON GITHUB
Every review,
with its code.
Each review will be published as its own public GitHub repository, linked to the exact arXiv version and the claims examined. Finished versions will be tagged as releases and archived with a citable DOI.
Code & results
Recorded inputs, the calculation itself, expected outputs and instructions to rerun it on your own machine.
Review status
Who has checked what (AI models, the project, an external specialist, the authors), what is still open, and what the result changes, if anything.
WHERE THINGS STAND
One researcher,
several AI models.
Today the reviews are done by one researcher working with several AI models. The models help re-derive equations and write and run code. The researcher checks and records each step, and all of the code goes into the published repository.
No result has yet been reviewed by an external specialist, and none is published here. Agreement between AI models is not treated as confirmation: every result carries a label saying how far it has been checked.
METHOD
Reproduce. Challenge. Review.
Recover the original result
Start with the paper’s own assumptions and data. Assessments using newer observations are recorded separately.
Test the explanation
Check conventions, analytic limits and numerical convergence. Use independent derivations or implementations where possible.
Get it reviewed
Ask a specialist in the field and the paper’s authors to respond before calling anything an error. Report reproducibility, scientific agreement and remaining uncertainty separately.
PUBLIC RESEARCH MATERIALS
Nothing released yet.
This catalogue will link each review to its GitHub repository, report and release. Confirmed results and error reports that turn out to be wrong will both be published.